CN117472478B

CN117472478B - Automatic flow generation method, related device and system, equipment and storage medium

Info

Publication number: CN117472478B
Application number: CN202311775435.9A
Authority: CN
Inventors: 刘杰; 黄鹏; 马永恒; 汤熙
Original assignee: iFlytek Co Ltd
Current assignee: iFlytek Co Ltd
Priority date: 2023-12-22
Filing date: 2023-12-22
Publication date: 2024-04-26
Anticipated expiration: 2043-12-22
Also published as: CN117472478A

Abstract

The application discloses an automatic flow generation method, and a related device, system, equipment and storage medium, wherein the automatic flow generation method comprises the following steps: acquiring a task description text of a target task; obtaining a plurality of subtasks forming a target task based on the task description text; for each subtask: generating an implementation flow of the subtask, operating the implementation flow of the subtask to obtain an operation result, responding to the success of the implementation flow operation, checking the operation result to obtain whether the characterization operation result accords with an expected check result, obtaining a determination result for characterizing whether the implementation flow of the subtask is successful or not based on the check result, and selecting whether to continue to execute the step of generating the implementation flow of the subtask for the next subtask based on the determination result until the implementation flow of each subtask is successfully generated to obtain the implementation flow for automatically executing the target task. By the aid of the scheme, the generation efficiency and quality of an automatic process can be improved.

Description

Automatic flow generation method, related device and system, equipment and storage medium

Technical Field

The present application relates to the field of artificial intelligence, and in particular, to an automated process generating method, and related apparatus, system, device, and storage medium.

Background

Thanks to the fact that the robot process automation (Robotic Process Automation, RPA) can simulate the manual operation mode of a user, the automation of the manual operation process of the user is realized, and the robot process automation is widely applied to various industries.

In the prior art, a user is usually required to manually arrange a flow, and then an automatic application program is created, so that the automation of the robot flow is realized. However, manual arrangement requires a lot of manpower and is limited by user arrangement experience, and the quality of the arrangement is difficult to be ensured. In view of this, how to improve the generation efficiency and quality of the automation flow is a problem to be solved.

Disclosure of Invention

The application mainly solves the technical problem of providing an automatic flow generating method, and a related device, a related system, a related device, a related storage medium and a related storage medium, and can improve the generating efficiency and the generating quality of an automatic flow.

In order to solve the above technical problem, a first aspect of the present application provides an automated process generating method, including: acquiring a task description text of a target task; obtaining a plurality of subtasks forming a target task based on the task description text; for each subtask: generating an implementation flow of the subtask, operating the implementation flow of the subtask to obtain an operation result, responding to the success of the implementation flow operation, checking the operation result to obtain whether the characterization operation result accords with an expected check result, obtaining a determination result for characterizing whether the implementation flow of the subtask is successful or not based on the check result, and selecting whether to continue to execute the step of generating the implementation flow of the subtask for the next subtask based on the determination result until the implementation flow of each subtask is successfully generated to obtain the implementation flow for automatically executing the target task.

In order to solve the above technical problem, a second aspect of the present application provides an automated process generating method, including: acquiring a task description text of a target task; obtaining a plurality of subtasks forming a target task based on the task description text; transmitting the subtasks to the flow generating agent; the process generation agent generates an implementation process of a subtask, and sends the implementation process of the subtask to the process execution agent, the implementation process of the process execution agent operates the subtask to obtain an operation result, and responds to the successful operation of the implementation process, sends the operation result to the execution check agent, executes the check agent to check the operation result, obtains whether the characterization operation result accords with an expected check result, and sends the check result to the cooperative hub agent; receiving a verification result from an execution verification agent, and obtaining a determination result indicating whether the realization flow of the subtask is successful or not based on the verification result; and selecting whether to return to the step of sending the subtasks to the flow generating agent or not based on the determination result until the realization flow of each subtask is successfully generated, so as to acquire the realization flow for automatically executing the target task.

In order to solve the above technical problem, a third aspect of the present application provides an automated process generating method, including: receiving subtasks from the cooperative hub agent; the cooperative hub agent acquires and analyzes a task description text of a target task to obtain a plurality of subtasks forming the target task, and sends the subtasks to the flow generation agent; generating an implementation flow of the subtasks, and sending the implementation flow of the subtasks to a flow execution agent; the method comprises the steps that a process execution agent runs an implementation process of a subtask to obtain a running result, the implementation process is responded to the success of the implementation process, the running result is sent to an execution check agent, the check agent is executed to check the running result to obtain whether the characterization running result accords with an expected check result, the check result is sent to a cooperative hub agent, the cooperative hub agent obtains a determination result whether the implementation process of the characterization subtask is successful or not based on the check result, and the step of sending the subtask to the process generation agent is selected based on the determination result until the implementation process of each subtask is successfully generated, so that the implementation process for automatically executing the target task is obtained.

In order to solve the above technical problem, a fourth aspect of the present application provides an automated process generating method, including: receiving an implementation flow of a subtask from a flow generation agent; the method comprises the steps that a cooperative hub agent obtains and analyzes a task description text of a target task to obtain a plurality of subtasks forming the target task, the subtasks are sent to a flow generation agent, the flow generation agent generates an implementation flow of the subtasks, and the implementation flow of the subtasks is sent to a flow execution agent; the realization flow of the operation subtask obtains an operation result; in response to successful implementation of the process operation, sending an operation result to an execution verification agent; and executing the verification operation result of the verification agent to obtain whether the characterization operation result accords with the expected verification result, sending the verification result to the cooperative hub agent, obtaining a determination result indicating whether the realization flow of the subtask is generated successfully based on the verification result by the cooperative hub agent, and selecting whether to return to the step of sending the subtask to the flow generating agent based on the determination result until the realization flow of each subtask is generated successfully so as to obtain the realization flow for automatically executing the target task.

In order to solve the above technical problem, a fifth aspect of the present application provides an automated process generating method, including: receiving an operation result from a flow execution agent; the method comprises the steps that a cooperative hub agent obtains and analyzes a task description text of a target task to obtain a plurality of subtasks forming the target task, the subtasks are sent to a flow generation agent, the flow generation agent generates an implementation flow of the subtasks, the implementation flow of the subtasks is sent to a flow execution agent, the implementation flow of the flow execution agent operates the subtasks to obtain an operation result, and the operation result is sent to an execution verification agent in response to the success of the implementation flow operation; checking the operation result to obtain whether the characterization operation result accords with an expected check result, and sending the check result to the cooperative hub intelligent agent; the cooperative hub agent obtains a determination result indicating whether the realization flow of the subtask is successful or not based on the verification result, and selects whether to return to the step of sending the subtask to the flow generation agent or not based on the determination result until the realization flow of each subtask is successfully generated, so as to obtain the realization flow for automatically executing the target task.

In order to solve the above technical problem, a sixth aspect of the present application provides an automated process generating method, including: the cooperative hub agent acquires a task description text of a target task, obtains a plurality of subtasks forming the target task based on the task description text, and sends the subtasks to the flow generation agent; the process generating agent receives the subtasks from the cooperative hub agent, generates an implementation process of the subtasks, and sends the implementation process of the subtasks to the process executing agent; the process execution agent receives the realization process of the subtasks from the process generation agent, runs the realization process of the subtasks to obtain a running result, and responds to the successful running of the realization process, and sends the running result to the execution verification agent; the execution verification agent receives the operation result from the flow execution agent, performs bulk verification on the operation result to obtain whether the characterization operation result accords with the expected verification result, and sends the verification result to the cooperative hub agent; the cooperative hub agent receives the verification result from the execution verification agent, obtains a determination result indicating whether the realization flow of the subtask is successful or not based on the verification result, and selects whether to return to the step of sending the subtask to the flow generation agent or not based on the determination result until the realization flow of each subtask is successfully generated, so as to obtain the realization flow for automatically executing the target task.

In order to solve the above-mentioned technical problem, a seventh aspect of the present application provides an electronic device, including a memory and a processor coupled to each other, where the memory stores program instructions, and the processor is configured to execute the program instructions to implement the automated process generating method in the first aspect to the fifth aspect.

In order to solve the above-mentioned technical problem, an eighth aspect of the present application provides an automated process generation system, including a cooperative hub agent, a process generation agent, a process execution agent, and an execution check agent that are communicatively connected to each other, where the cooperative hub agent is configured to execute the automated process generation method of the first aspect, the process generation agent is configured to execute the automated process generation method of the second aspect, the process execution agent is configured to execute the automated process generation method of the third aspect, and the execution check agent is configured to execute the automated process generation method of the fourth aspect.

In order to solve the above technical problem, a ninth aspect of the present application provides a computer-readable storage medium storing program instructions executable by a processor for implementing the automated process generating methods of the first to fifth aspects.

According to the scheme, the task description text of the target task is obtained, a plurality of subtasks forming the target task are obtained based on the task description text, the realization flow of the subtasks is generated, the realization flow of the subtasks is operated to obtain the operation result, when the realization flow is operated successfully, the operation result is checked to obtain whether the characterization operation result accords with the expected check result, the determination result indicating whether the realization flow of the subtasks is successfully generated is obtained based on the check result, whether the step of generating the realization flow of the subtasks is continuously executed for the next subtask is selected based on the determination result until the realization flow of each subtask is successfully generated, so that the realization flow for automatically executing the target task is obtained, therefore, the automatic generation of the flow for realizing the target task can be triggered based on the task description text about the target task, and the generation quality of the realization flow is judged based on the operation result and the determination result, and the generation efficiency and quality of the automatic flow can be improved.

Drawings

FIG. 1 is a flow diagram of an embodiment of an automated flow generation method of the present application;

FIG. 2 is a flow diagram of an embodiment of an automated flow generation method of the present application;

FIG. 3 is a flow chart of an embodiment of an automated flow generation method of the present application;

FIG. 4 is a flow chart of an embodiment of an automated flow generation method of the present application;

FIG. 5 is a flow chart of an embodiment of an automated flow generation method of the present application;

FIG. 6 is a flow diagram of an embodiment of an automated flow generation method of the present application;

FIG. 7 is a schematic diagram of an automated process generating apparatus according to one embodiment of the present application;

FIG. 8 is a schematic diagram of an automated process generating apparatus according to one embodiment of the present application;

FIG. 9 is a schematic diagram of an automated process generating apparatus according to one embodiment of the present application;

FIG. 10 is a schematic diagram of an automated process generating apparatus according to one embodiment of the present application;

FIG. 11 is a schematic diagram of an automated process generating apparatus according to one embodiment of the present application;

FIG. 12 is a schematic diagram of an automated process generating apparatus according to one embodiment of the present application;

FIG. 13 is a schematic diagram of a frame of an embodiment of an electronic device of the present application;

FIG. 14 is a schematic diagram of a framework of one embodiment of an automated process generation system of the present application;

FIG. 15 is a schematic diagram of a frame of an embodiment of a computer readable storage medium of the present application.

Detailed Description

The following describes embodiments of the present application in detail with reference to the drawings.

In the following description, for purposes of explanation and not limitation, specific details are set forth such as the particular system architecture, interfaces, techniques, etc., in order to provide a thorough understanding of the present application.

The terms "system" and "network" are often used interchangeably herein. The term "and/or" is herein merely an association relationship describing an associated object, meaning that there may be three relationships, e.g., a and/or B, may represent: a exists alone, A and B exist together, and B exists alone. In addition, the term "/" herein generally indicates that the associated object is an "or" relationship. Further, "a plurality" herein means two or more than two.

Referring to fig. 1, fig. 1 is a flow chart illustrating an embodiment of an automated flow generation method according to the present application. Specifically, the method may include the steps of:

Step S11: and acquiring a task description text of the target task.

In the embodiment of the disclosure, the task description text describes the target task in a natural language, which may be input by a user, and is exemplified by "i want to purchase a pair of black basketball shoes on the a platform" or "watch documentary C using the B software", which is not exemplified here.

Step S12: based on the task description text, a plurality of subtasks constituting the target task are obtained.

In one implementation scenario, based on the semantics of the task description text and the implementation logic of the target task, a plurality of subtasks are obtained, for example, the task description text is "i want to purchase a pair of black basketball shoes on the platform a", the target task is sequentially decomposed into subtask 1 "open browser", subtask 2 "locate input frame", subtask 3 "input a platform website", subtask 4 "locate search frame", subtask 5 "input 'black basketball shoes'", and subtask 6 "click search button", so as to obtain a plurality of subtasks forming the target task. According to the method, the target task is decomposed into the plurality of subtasks which are as reasonable as possible based on the processing logic of divide and conquer, and the efficiency of automatic flow generation is improved.

Step S13: for each subtask: generating an implementation flow of the subtask, operating the implementation flow of the subtask to obtain an operation result, responding to the success of the implementation flow operation, checking the operation result to obtain whether the characterization operation result accords with an expected check result, obtaining a determination result for characterizing whether the implementation flow of the subtask is successful or not based on the check result, and selecting whether to continue to execute the step of generating the implementation flow of the subtask for the next subtask based on the determination result until the implementation flow of each subtask is successfully generated to obtain the implementation flow for automatically executing the target task.

In one implementation scenario, based on the description document of each function in the flow library, a function matched with the subtask is selected as a target function, and the target function is called to generate an implementation flow of the subtask. According to the method, the target function is called based on the description document of the function, so that the efficiency and the quality of the process generation are improved.

In a specific implementation scenario, a description document of a function is composed by a developer, parameters can be transferred to the function when the function is called based on the description document, the parameters are used as input of the function in a function body, a result can be returned after the function is executed, and the multiplexing of the flow is realized through the call of the function.

In a specific implementation scenario, a developer may modify and adjust a description document about a function, and exemplarily modify a function name, use a natural language text that is clearly and accurately expressed in language, and avoid ambiguity of the function name, where a certain difference exists between names of function functions of different functions.

In one implementation scenario, the implementation flow of the subtask is run, and a running result is obtained, wherein the running result comprises one of a running result representing a running failure and a running result representing a running success.

In a specific implementation scenario, when the implementation process operation is abnormal and the operation result represents that the implementation process operation fails, the implementation process of the subtask is regenerated based on the error reporting information of the implementation process operation failure of the task in the operation result, and the error reporting information includes, but is not limited to: violating grammar rules of programming language, null pointer exception, type conversion exception, etc., and not being exemplified one by one, the realization flow of subtasks is regenerated by combining error reporting information.

In another specific implementation scenario, when the implementation process of the subtask is generated by selecting and calling a function based on the description document of each function in the process library, and the generated implementation process fails to run, the description document of the function in the process library is corrected based on error reporting information, which includes, but is not limited to: violating grammar rules of programming languages, null pointer exceptions, type conversion exceptions and the like, and not being exemplified one by one, correcting the description document of the function in the flow library by combining error reporting information, reducing the error rate of subsequent function call and improving the generation efficiency and quality of an automatic flow.

In one implementation scenario, when the operation result characterizes the implementation process to operate successfully, the operation result is checked to obtain whether the characterization operation result accords with an expected check result, and based on the check result, a determination result whether the realization process of the characterization subtask generates a success is obtained, and one of the determination result of the successful generation of the characterization result and the determination result of the failed generation of the characterization is determined.

In one implementation scenario, the detection determination result characterizes success or failure of generation, and based on the determination result, a step of whether to continue to execute the implementation flow of the generation subtask for the next subtask is selected until the implementation flow of each subtask is successfully generated, so as to obtain the implementation flow for automatically executing the target task.

In a specific implementation scenario, when the verification result represents the operation result and accords with the expectation, a determination result that the realization flow of the representation subtask is successfully generated is obtained, and when the verification result represents the operation result and does not accord with the expectation, a determination result that the realization flow of the representation subtask is failed to be generated is obtained.

In a specific implementation scenario, when the determination result indicates that the generation is successful, the steps of the implementation flow for generating the subtasks are continuously executed for the next subtask. And when determining that the result representation fails to be generated, acquiring an analysis result of whether the representation target task is reasonably decomposed into a plurality of subtasks, and returning to a text based on task description based on the analysis result to acquire a plurality of subtasks constituting the target task.

In a specific implementation scenario, when the analysis result characterizes that the target task is reasonably decomposed into a plurality of subtasks, a plurality of subtasks with better quality cannot be regenerated based on the task description text, an opinion description text of the target object decomposing the target task can be obtained, the target object is an human manager, the opinion description text is a natural language text, the opinion description text is combined, the step of returning to the subtasks based on the task description text to obtain a plurality of subtasks forming the target task is performed, and a new plurality of subtasks are regenerated. According to the method, auxiliary information is provided for sub-task re-acquisition based on the opinion description text of the human manager, and the generation quality of the sub-tasks is improved.

In another specific implementation scenario, when the analysis result characterizes that the target task is not reasonably decomposed into a plurality of subtasks, a plurality of subtasks with better quality can be regenerated based on the task description text, the step of obtaining a plurality of subtasks forming the target task based on the task description text is returned, and a plurality of subtasks with better quality are regenerated.

Referring to fig. 2, fig. 2 is a flow chart illustrating an embodiment of an automated flow generation method according to the present application. It should be noted that, the embodiments of the present disclosure mainly describe differences from the foregoing disclosed embodiments, and the same or similar points as those of the foregoing disclosed embodiments may be referred to the related descriptions in the foregoing disclosed embodiments, which are not repeated herein. Specifically, the embodiment of the disclosure uses the cooperative pivot agent as an execution main body, and may include the following steps:

Step S21: and acquiring a task description text of the target task.

In one implementation scenario, a collaboration hub Agent obtains task description text about a target task input by a user, where the Agent (Agent) is an entity residing in an environment, and may interpret data obtained from the environment and reflect events occurring in the environment, and perform actions affecting the environment, where it is noted that each Agent involved in the present application may be either hardware or software, and specific principles may refer to technical details of the Agent and are not described herein.

Step S22: based on the task description text, a plurality of subtasks constituting the target task are obtained.

In one implementation scenario, the collaborative hub agent decomposes based on the semantics of the task description text and the implementation logic of the target task to obtain a plurality of subtasks, for example, the task description text is "i want to purchase a pair of black basketball shoes on the a platform", the target task is decomposed into subtask 1 "open browser", subtask 2 "locate input frame", subtask 3 "input a platform website", subtask 4 "locate search frame", subtask 5 "input 'black basketball shoes'", and subtask 6 "click search button" in sequence, so as to obtain a plurality of subtasks constituting the target task. According to the method, the target task is decomposed into the plurality of subtasks which are as reasonable as possible based on the divide-and-conquer processing logic, interaction of multiple agents and execution of the target task are achieved, and efficiency of automatic flow generation is improved.

In a specific implementation scenario, the collaboration hub agent contains a first large model, which is a large language model, refers to a deep learning model trained using a large amount of text data, and can generate natural language text or understand meaning of the language text, and incorporates artificial intelligence models of various language knowledge and language rules. The large language model can process various natural language tasks and has the capability of understanding, generating and processing natural language. And performing task decomposition on the task description text based on the first large model to obtain a plurality of subtasks forming the target task. According to the method, the task description text is decomposed by using the first large model, so that a plurality of subtasks are obtained, the decomposition rationality of the target task is improved, and the quality of automatic flow generation is improved as much as possible.

It should be noted that the first large model may include, but is not limited to: the network architecture of the first large model is not limited herein by the open source large model such as LLAMA, bloom, etc.

In a specific implementation scenario, the first large model is obtained by performing parameter trimming based on a preset large model, and exemplary manners of trimming parameter trimming may include, but are not limited to: LORA, PT, SSF, etc., the specific manner of fine tuning the parameters is not limited herein. Taking the LORA for parameter tuning as an example, a network structure similar to the encoder-decoder architecture is added between the input and the output of the preset large model, and the network structure has only a small amount of parameters, so that the network structure is executed in parallel with the original path of the preset large model, and the characteristic distribution can be migrated to the desired distribution by only adjusting a small amount of parameters through the parameter tuning.

In a specific implementation scenario, the first large model performs parameter fine adjustment in advance based on human feedback Reinforcement Learning (RLHF), and a human manager provides assistance to parameter adjustment of the first large model by providing additional rewards and punishment signals or data required for training the rewards model, so that the first large model after performing parameter fine adjustment can stably generate reasonable and detailed subtasks according to a target task, and specific principles can refer to technical details of human feedback reinforcement learning and are not described herein. The following formula can be referred to specifically:

in the formula (1), Representing decomposed subtasks,/>Representing a first large model,/>Parameters representing the first large model,/>Representing task description text. The formula (1) characterizes the first large model, a plurality of subtasks are obtained based on the decomposition of task description text, human feedback reinforcement learning is introduced to carry out parameter fine adjustment on the first large model, and the following formula can be specifically referred to:

In the formula (2), Representing decomposed subtasks,/>Representing a first large model,/>Parameters representing the first large model,/>Representing task description text,/>Representing the artificial feedback signal. The formula (2) characterizes that the first large model after the artificial feedback signal is introduced to carry out parameter fine adjustment on the first large model is decomposed based on the task description text to obtain a plurality of subtasks, and the first large model after the parameter fine adjustment can be concretely referred to the following formula:

In the formula (3), Representing a first large model,/>Parameters representing the first large model,/>Representing task description text,/>Representing artificial feedback signal,/>Representing the minimum parameter trim loss. Equation (3) characterizes the first large model after parameter fine tuning.

In a specific implementation scenario, auxiliary information is provided for obtaining a subtask based on a front-back dialogue text of a task description text of a target task, and the following formula can be referred to specifically:

In the formula (4) of the present invention, Representing decomposed subtasks,/>Representing a first large model,/>Parameters representing the first large model,/>Representing task description text,/>Representing the front and back dialog text. The formula (4) characterizes the first large model to obtain a plurality of subtasks based on the task description text and the front and back dialogue text, the front and back dialogue text provides auxiliary information for the logic understanding of the first large model to the target task, and the generation quality of the subtasks is improved.

In one particular implementation scenario, training the bonus model to assist in fine-tuning parameters of the first large model, exemplary model types of the bonus model may include, but are not limited to: the network architecture of the rewards model is not limited herein for the LLAMA, bloom, etc. open source large model. The method comprises the steps of obtaining a sample data set, wherein sample data in the sample data set is a sample description text of a sample task, generating a plurality of subtask sets comprising a plurality of sample subtasks based on the sample description text, carrying out quality scoring on the quality of each subtask set by a human evaluation personnel based on human preference, marking each subtask set and corresponding quality score in the sample description text, and obtaining a reward model for human feedback reinforcement learning by an exemplary way, wherein the higher the score is used for representing that the corresponding subtask set is more in line with the human preference, the lower the score is used for representing that the corresponding subtask set is less in line with the human preference, carrying out predictive scoring on each subtask set of each sample description text in the sample data set based on the reward model, and adjusting model parameters of the reward model based on the difference between the predictive scoring and the marked quality score until the reward model meets a first preset condition. It should be noted that the difference may be measured using a loss function such as cross entropy. According to the method, the reward model provides assistance for parameter fine adjustment of the first large model, parameter adjustment is carried out on the reward model based on the sample data set, evaluation quality of the reward model is improved as much as possible, effective auxiliary information is provided for parameter fine adjustment of the first large model, and therefore generation quality of automatic flow generation can be improved.

In one specific implementation scenario, the first preset condition may include, but is not limited to: the number of training is not less than the number threshold, the training loss is less than the loss threshold, and the like, and is not limited herein.

In a specific implementation scenario, the reward model is a sub-model of a first large model, the first large model adjusts the generation bias and the generation weight, the sample data set generates a plurality of subtask sets comprising a plurality of sample subtasks based on the first large model, the human evaluator performs quality scoring on the quality of each subtask set based on human preference, and each subtask set and the corresponding quality score are respectively marked in the sample description text. According to the method, the sample description text is generated by a plurality of subtask sets based on the first large model, so that the quality of auxiliary information for parameter fine adjustment of the reward model for the first large model is improved as much as possible, and the generation quality of automatic flow generation can be improved.

Step S23: and sending the subtasks to the flow generating agent.

In one implementation scenario, the cooperative hub agent sends the subtasks as a single-threaded task sending state, that is, sends each subtask in turn based on the sequence of a plurality of subtasks, and sends the next subtask to the generating agent only after the current subtask successfully generates the implementation flow, for example, the cooperative hub agent sequentially decomposes the target task into a subtask 1, a subtask 2, a subtask 3, a subtask 4, a subtask 5 and a subtask 6 to obtain a plurality of subtasks constituting the target task, according to the implementation logic of the target task, firstly sends the subtask 1 to the flow generating agent, and sends the subtask 2 to the flow generating agent after the cooperative hub agent determines that the subtask 1 successfully generates the implementation flow, so as to push the following.

In another implementation scenario, the cooperative hub agent sends subtasks to be in a multithreaded task sending state, it can be understood that other agents also process subtask data based on multithreading, that is, multiple subtasks respectively perform data processing in different threads, for example, the cooperative hub agent sequentially breaks down a target task into subtask 1, subtask 2, subtask 3, subtask 4, subtask 5 and subtask 6, the intelligent parsing task body has thread A1, thread A2 and thread A3, the flow generating agent has thread B1, thread B2 and thread B3, uses thread A1 to send subtask 1 to thread B1, uses thread A2 to send subtask 2 to thread B2, uses thread A3 to send subtask 3 to thread B3, and so on, which is not exemplified here.

In a specific implementation scenario, when the cooperative hub agent sends a subtask to be in a multi-thread task sending state, each subtask obtained based on a target task decomposition sequence is marked with a generation sequence number, an implementation flow corresponding to each subtask is marked with a generation sequence number consistent with the subtask, and splicing and combining are performed based on the implementation flow generated by each subtask and the generation sequence number, so that an implementation flow for automatically executing the target task is obtained.

In one implementation scenario, after the cooperative hub agent sends the subtasks to the flow generation agent, the flow generation agent generates an implementation flow of the subtasks, and sends the implementation flow to the flow execution agent, the flow execution agent runs the implementation flow of the subtasks to obtain a running result, and under the condition that the implementation flow is successful in running, the running result is sent to the execution verification agent, the execution verification agent is used for verifying whether the running result accords with the expected or not, and the verification result representing whether the running result accords with the expected or not is sent to the cooperative hub agent, so that interaction of multiple agents can be triggered based on a task description text about the target task, flow automatic generation of the target task is achieved, and universality and generation quality of automatic flow generation are improved.

Step S24: and receiving a verification result from the execution verification agent, and obtaining a determination result indicating whether the realization flow of the subtask is successful or not based on the verification result.

In one implementation scenario, performing the verification results of the verification agent includes characterizing that the operational result meets an expected first verification result and characterizing that the operational result does not meet an expected second verification result, the reasons for the operational result not meeting the expected include, but are not limited to, a function call error, a subtask semantic understanding error, and the like. And obtaining a determination result indicating whether the realization flow of the subtask is successful or not based on the verification result, wherein the determination result indicates that the generation is successful when the verification result is a first verification result, and the determination result indicates that the generation is failed when the verification result is a second verification result.

In a specific implementation scenario, when the verification result represents that the operation result meets the expectation, a determination result that the realization flow of the representation subtask is successfully generated is obtained.

In another specific implementation scenario, when the verification result ensures that the operation result does not meet the expectations, a determination result indicating that the implementation flow generation of the subtask fails is obtained.

Step S25: and selecting whether to return to the step of sending the subtasks to the flow generating agent or not based on the determination result until the realization flow of each subtask is successfully generated, so as to acquire the realization flow for automatically executing the target task.

In one implementation scenario, whether the determination result representation is generated successfully is detected, when the determination result representation subtask implementation flow is generated successfully, a step of sending the subtask to the flow generation agent is returned, and a subsequent step is carried out by sending the next subtask until the implementation flow of each subtask is generated successfully, so that the implementation flow for automatically executing the target task is obtained.

In another implementation scenario, whether the determination result representation is successfully generated is detected, when the determination result representation subtask implementation flow fails to be generated, an analysis result that whether the representation target task is decomposed into a plurality of subtasks is obtained, a step of obtaining a plurality of subtasks forming the target task based on the analysis result is returned based on the task description text, a plurality of new subtasks are obtained through re-decomposition, and a subsequent step is executed until the realization flow of each subtask is successfully generated, so that the realization flow for automatically executing the target task is obtained.

In a specific implementation scenario, when the analysis result characterizes that the target task is decomposed into a plurality of subtasks reasonably, the first large model cannot generate a plurality of subtasks with better quality based on a network structure of the first large model, an opinion description text of the target object decomposing the target task is obtained, the target object is an human manager, the opinion description text is a natural language text, the first large model is combined with the opinion description text, the step of returning to the subtasks forming the target task based on the task description text is performed, and a plurality of new subtasks are regenerated. According to the method, auxiliary information is provided for the re-acquisition of the subtasks of the first large model based on the opinion description text of the human manager, and the generation quality of the subtasks is improved.

In another specific implementation scenario, when the analysis result characterizes that the target task is not reasonably decomposed into a plurality of subtasks, returning to a step of obtaining a plurality of subtasks constituting the target task based on the task description text, and regenerating a plurality of subtasks with better new quality.

It should be noted that, the method for obtaining the analysis result is not limited in the present application, for example, a task analysis model is constructed, and based on the task analysis model, an analysis result indicating whether a plurality of subtasks are reasonable is obtained, and a network result of the task analysis model is not limited, which may include, but is not limited to: LLAMA, bloom, etc. open source large models.

It should be noted that, as described above, the cooperative hub agent is not only responsible for understanding the task description text, but also responsible for decomposing the target task into a plurality of subtasks as reasonable as possible to solve the problem by dividing the thought, and the cooperative hub agent is also responsible for optimizing the subtasks based on the verification result fed back by the execution verification agent, the opinion description text of the target object, and the like. It can be seen that it acts as a key hub in the overall link generated by the automated process, and is therefore referred to as a cooperative hub agent.

According to the scheme, the cooperative hub agent acquires the task description text of the target task, acquires a plurality of subtasks forming the target task based on the task description text, transmits the subtasks to the flow generation agent, the flow generation agent receives the subtasks from the cooperative hub agent, generates an implementation flow of the subtasks, transmits the implementation flow of the subtasks to the flow execution agent, and after the flow execution agent receives the implementation flow of the subtasks from the flow generation agent, operates the implementation flow of the subtasks to obtain an operation result, and responds to the successful implementation flow, transmits the operation result to the execution verification agent, executes the verification agent to receive the operation result from the flow execution agent, and then verifies the operation result to obtain a representation operation result which accords with an expected verification result, and transmits the verification result to the cooperative hub agent.

Referring to fig. 3, fig. 3 is a flow chart illustrating an embodiment of an automated flow generation method according to the present application. It should be noted that, the embodiments of the present disclosure mainly describe differences from the foregoing disclosed embodiments, and the same or similar points as those of the foregoing disclosed embodiments may be referred to the related descriptions in the foregoing disclosed embodiments, which are not repeated herein. Specifically, the embodiment of the disclosure uses the program generation agent as an execution subject, and may include the following steps:

Step S31: sub-tasks from the collaborative hub agent are received.

In one implementation scenario, the program generating Agent receives subtasks from the cooperative hub Agent, where the Agent (Agent) is an entity residing in the environment, and can interpret data obtained from the environment and reflect events occurring in the environment, and perform actions affecting the environment.

In an implementation scenario, the cooperative hub agent acquires and analyzes the task description text of the target task to obtain a plurality of subtasks forming the target task, so as to send the subtasks to the flow generation agent, and the cooperative hub agent may generate and send the subtasks, which are described in detail in the above embodiments, and are not described herein for brevity.

Step S32: and generating the realization flow of the subtasks and sending the realization flow of the subtasks to a flow execution agent.

In one implementation scenario, an implementation flow of a subtask is generated based on API (Application Programming Interface ) specification documents, each API specification document comprising: an API name, a parameter list, an API description, etc. (as shown in table 1 below), the API name providing a digest of the API to facilitate linking tasks to this API to call an objective function, the parameter list of the API including input parameters and return values, each parameter having a parameter name, a parameter description, a data type, and default values, the API description containing more information about the function, manner of operation, input and output, and possible errors or anomalies of the API, in some scenarios, providing a use case for a complex application program interface for demonstrating how to use the application program interface, or providing a combined description for guiding a model how to complete complex user instructions in connection with multiple APIs.

TABLE 1

In a specific implementation scenario, based on the description document of each function in the flow library, a function matched with the subtask is selected as a target function, and the target function is called to generate an implementation flow of the subtask. According to the method, the target function is called based on the description document of the function, so that the efficiency and the quality of the process generation are improved.

In a specific implementation scenario, a description document of a function is composed by a developer, parameters can be transferred to the function when the function is called based on the description document, the parameters are used as input of the function in a function body, a result can be returned after the function is executed, and the function is reused by calling the function.

In another specific implementation scenario, the process generation agent contains a second large model, which may include, but is not limited to: the network architecture of the second large model is not limited herein by the open source large model of LLAMA, bloom, etc. And generating an explanation document for obtaining the function based on the implementation flow of the second large model processing function. According to the method, the second large model is used for generating the description document of the function, and the generation efficiency of the description document is improved on the premise that the generation quality of the description document is ensured as much as possible.

In one implementation scenario, after sending an implementation flow of a subtask to a flow execution agent, the flow execution agent runs the implementation flow of the subtask to obtain a running result, and in response to the implementation flow running success, sends the running result to an execution check agent, executes the check agent check running result to obtain a check result which indicates whether the running result accords with expectations, and sends the check result to a cooperative hub agent, the cooperative hub agent obtains a determination result which indicates whether the implementation flow of the subtask is successful based on the check result, and selects whether to return to the step of sending the subtask to the flow generation agent based on the determination result until the implementation flow of each subtask is successfully generated, so as to obtain the implementation flow for automatically executing the target task.

In a specific implementation scenario, after the implementation process of the subtask is sent to the process execution agent, an operation result sent by the process execution agent in response to the implementation process operation failure is received, the operation result includes error reporting information of the implementation process operation failure of the subtask, and the error reporting information is combined with the error reporting information to return to a step of generating the implementation process of the subtask, and the implementation process of the subtask is regenerated based on the error reporting information. Exemplary error reporting information includes, but is not limited to: violating grammar rules of the programming language, null pointer exceptions, type conversion exceptions, etc., are not exemplified herein.

According to the scheme, the cooperative hub agent acquires the task description text of the target task, acquires a plurality of subtasks forming the target task based on the task description text, transmits the subtasks to the flow generation agent, the flow generation agent receives the subtasks from the cooperative hub agent, generates an implementation flow of the subtasks, transmits the implementation flow of the subtasks to the flow execution agent, and after the flow execution agent receives the implementation flow of the subtasks from the flow generation agent, operates the implementation flow of the subtasks to obtain an operation result, responds to the successful implementation flow, transmits the operation result to the execution verification agent, executes the verification agent to receive the operation result from the flow execution agent, verifies the operation result, obtains a representation operation result to conform to an expected verification result, transmits the verification result to the cooperative hub agent, receives the verification result from the execution verification agent, obtains a determination result for judging whether the implementation flow of the subtasks is successful or not based on the verification result, and selects whether to return to the step of the flow generation agent until the realization of each subtask is successful, so that the realization flow of the subtasks can be automatically triggered based on the task description related to the target task, the automatic interactive description is realized, the quality of the realization flow of the tasks is improved, and the automatic generation of the realization flow of the tasks is realized, and the quality of the realization of the automatic interactive description is achieved.

Referring to fig. 4, fig. 4 is a flow chart illustrating an embodiment of an automated flow generation method according to the present application. It should be noted that, the embodiments of the present disclosure mainly describe differences from the foregoing disclosed embodiments, and the same or similar points as those of the foregoing disclosed embodiments may be referred to the related descriptions in the foregoing disclosed embodiments, which are not repeated herein. Specifically, the embodiment of the disclosure uses a flow execution agent as an execution subject, and may include the following steps:

Step S41: and receiving the realization flow of the subtasks from the flow generation agent.

In one implementation scenario, a process execution Agent receives an implementation process from a process generation Agent, where the Agent (Agent) is an entity residing in an environment, and may interpret data obtained from the environment and reflect an event occurring in the environment, and perform an action affecting the environment.

In one implementation scenario, before receiving an implementation flow of a subtask from a flow generation agent, a flow execution agent acquires and analyzes a task description text of a target task in cooperation with a hub agent to obtain a plurality of subtasks constituting the target task, to send the subtasks to the flow generation agent, the flow generation agent generates an implementation flow of the subtasks, and sends the implementation flow of the subtasks to the flow execution agent. Reference should be made specifically to the foregoing detailed description of the embodiments of the present disclosure, which is not repeated here for brevity.

Step S42: and the realization flow of the operation subtask obtains an operation result.

In one implementation scenario, the implementation process of the process execution agent execution subtasks obtains an operation result, where the operation result includes a first operation result indicating that the operation is successful and a second operation result indicating that the operation is failed. The implementation flow may be specifically described in the foregoing disclosure embodiments, and for brevity, details are not repeated herein.

Step S43: and in response to successful flow operation, sending an operation result to the execution verification agent.

In one implementation scenario, after the implementation process is successfully operated, an operation result is sent to an execution check agent, the execution check agent checks the operation result, whether the characterization operation result accords with an expected check result is obtained, the check result is sent to a cooperative hub agent, the cooperative hub agent obtains a determination result whether the implementation process of the characterization subtask is successfully generated based on the check result, and whether the subtask is returned to the process generation agent step is selected based on the determination result until the implementation process of each subtask is successfully generated, so that the implementation process for automatically executing the target task is obtained.

In another implementation scenario, when the implementation process fails, the process execution agent sends an operation result to the process generation agent, and the operation result includes error reporting information of the sub-task implementation process failure, where the error reporting information includes, but is not limited to: violating grammar rules of the programming language, null pointer exceptions, type conversion exceptions, etc., are not exemplified herein. And the process generation agent combines the error reporting information to regenerate the realization process of the subtasks.

In one implementation scenario, when the implementation flow fails to run, a running result is sent to the document correction agent, and the running result contains error reporting information of the implementation flow failure of the subtasks, and exemplary error reporting information includes, but is not limited to: violating grammar rules of the programming language, null pointer exceptions, type conversion exceptions, etc., are not exemplified herein. The realization flow of the subtasks is generated by a flow generating agent selecting and calling the function based on the description document of the function in the flow library, and a document correction agent corrects the description document based on the error reporting information.

In one specific implementation scenario, when the error reporting information in the document correction agent meets a second preset condition, the description document is corrected based on the error reporting information. Exemplary second preset conditions include, but are not limited to, not less than a preset number of error reporting information, not less than a preset frequency of occurrence of the same error reporting information, etc., which are not exemplified herein.

In a specific implementation scenario, the quality of the description document of the function is judged based on the error reporting information, and the description document of each function is evaluated and scored based on an evaluation model or the description document of each function is evaluated and scored based on manual combination of the error reporting information, wherein the more the error reporting information of the same function is, the lower the quality of the corresponding description document is represented, and conversely, the less the error reporting information of the same function is, and the better the quality of the corresponding description document is represented. It should be noted that the model type of the evaluation model is not limited in the present application, and may include, but is not limited to: LLAMA, bloom, etc., the network architecture of the evaluation model is not limited herein.

In a specific implementation scenario, a developer accesses a document correction agent to obtain error reporting information, and gathers the error reporting information for the developer to adjust a description document of a function.

Referring to fig. 5, fig. 5 is a flow chart illustrating an embodiment of an automated flow generation method according to the present application. It should be noted that, the embodiments of the present disclosure mainly describe differences from the foregoing disclosed embodiments, and the same or similar points as those of the foregoing disclosed embodiments may be referred to the related descriptions in the foregoing disclosed embodiments, which are not repeated herein. Specifically, the embodiment of the disclosure uses the execution verification agent as an execution subject, and may include the following steps:

step S51: and receiving an operation result from the flow execution agent.

In one implementation scenario, the execution verification Agent receives the operation result from the flow execution Agent, where the Agent (Agent) is an entity residing in the environment, and may interpret data obtained from the environment and reflect an event occurring in the environment, and perform an action affecting the environment.

In one implementation scenario, before an execution verification agent receives an operation result from a flow execution agent, a cooperative hub agent obtains and analyzes a task description text of a target task to obtain a plurality of subtasks constituting the target task, so as to send the subtasks to a flow generation agent, the flow generation agent generates an implementation flow of the subtasks, and sends the implementation flow of the subtasks to the flow execution agent, the implementation flow of the flow execution agent operates the subtasks to obtain an operation result, and in response to the implementation flow operation success, the operation result is sent to the execution verification agent. Reference should be made specifically to the foregoing detailed description of the embodiments of the present disclosure, which is not repeated here for brevity.

Step S52: and checking the operation result to obtain whether the characterization operation result accords with the expected check result, and sending the check result to the cooperative hub intelligent agent.

In one implementation scenario, the verification results include a first verification result that characterizes the operational result as expected and a second verification result that characterizes the operational result as not expected. Reasons for successful execution of the process but not expected include, but are not limited to, mismatching of the objective function with the subtasks, unclear subtask semantics, etc., which are not exemplified here.

In one implementation scenario, after sending the verification result to the cooperative hub agent, the cooperative hub agent obtains a determination result indicating whether the implementation flow of the subtask is successful based on the verification result, and selects whether to return to the step of sending the subtask to the flow generation agent based on the determination result until the implementation flow of each subtask is successfully generated, so as to obtain the implementation flow for automatically executing the target task. Reference should be made specifically to the foregoing detailed description of the embodiments of the present disclosure, which is not repeated here for brevity.

Referring to fig. 6, fig. 6 is a flow chart illustrating an embodiment of an automated flow generation method according to the present application. Specifically, each agent in the embodiment of the present disclosure is disposed in the same hardware environment or the same software environment, and may include the following steps:

Step S61: the cooperative hub agent acquires a task description text of the target task, obtains a plurality of subtasks constituting the target task based on the task description text, and sends the subtasks to the flow generation agent.

Reference may be made specifically to the foregoing disclosed embodiments, and details are not repeated here.

Step S62: the process generating agent receives the subtasks from the cooperative hub agent, generates an implementation process of the subtasks, and sends the implementation process of the subtasks to the process executing agent.

Step S63: the process execution agent receives the realization process of the subtasks from the process generation agent, runs the realization process of the subtasks to obtain a running result, and responds to the successful running of the realization process, and sends the running result to the execution verification agent.

Step S64: the execution verification agent receives the operation result from the process execution agent, verifies the operation result to obtain a verification result which indicates whether the operation result accords with the expectation, and sends the verification result to the cooperative hub agent.

Step S65: the cooperative hub agent receives the verification result from the execution verification agent, obtains a determination result indicating whether the realization flow of the subtask is successful or not based on the verification result, and selects whether to return to the step of sending the subtask to the flow generation agent or not based on the determination result until the realization flow of each subtask is successfully generated, so as to obtain the realization flow for automatically executing the target task.

In the embodiment of the present disclosure, each agent involved in the present disclosure may be hardware or software, and specific principles may refer to technical details of the agent, which are not described herein.

In one implementation scenario, flow generation is implemented through interactive conversations of multiple agents, and illustratively, a display window is displayed to a user, after the cooperative hub agent acquires a task description text "target task" of a target task, each of the obtained sub-tasks "sub-task 1", "sub-task 2", "sub-task 3", "sub-task 4", "sub-task 5", and "sub-task 6" is displayed in the display window, and the sub-tasks are sent to a flow generation agent, for example, the cooperative hub agent sends "sub-task 1" to the flow generation agent at the display window, the flow generation agent generates an implementation flow "code 1" and displays the implementation flow "code 1" at the display window after receiving the sub-task 1, and the flow execution agent sends "code 1" to the flow execution agent at the display window after receiving the "code 1", displaying the running process of the code 1 in a display window, displaying the running result, when the process running is successful, displaying the running result in the display window to be sent to an execution check agent, checking the running result after the execution check agent receives the running result, obtaining whether the characterization running result accords with the expected check result, displaying the check result in the display window and displaying the sending check result to the cooperative hub agent, obtaining a determination result for characterizing whether the realization flow of the subtask is successful or not based on the check result after the cooperative hub agent receives the check result from the execution check agent, displaying the determination result in a display page, and displaying the step of selecting whether to send the subtask back to the flow generation agent based on the determination result in the display window until the realization flow of each subtask is successfully generated, to obtain an implementation flow for automatically executing the target task. The method can strengthen the generation process of the user perception automation flow and timely follow up the generation of the automation flow. It should be noted that, the present application does not limit the form of interactive display among the agents, and the above specific embodiment is only one possible implementation manner, which is not described here again.

In one particular implementation scenario, a user may view a presentation window in real-time and perform actions of pausing, modifying, etc. a generation process of an automated process based on related content displayed in the presentation window. For example, after the implementation process is generated, the process generation agent displays the implementation process on the display window, and the user perceives the implementation process based on his own knowledge, experience, etc., and when the user considers that there is a modifiable error in the implementation process, the modification can be performed based on the visual content. It can be appreciated that the above specific embodiment is only one possible implementation manner, and the visual content in the automatic process generation can achieve interactive participation of the user, which is not described herein. According to the method, the automatic flow generation is visualized, the user can conveniently check the automatic flow, interaction between the automatic flow generation and the user is realized, and the generation efficiency of the automatic flow can be improved under the condition that the user has relevant knowledge and experience.

Referring to fig. 7, fig. 7 is a schematic diagram of a frame of an embodiment of an automatic process generating apparatus 70 according to the present application, where the automatic process generating apparatus 70 includes a task obtaining module 71, a task disassembling module 72, and a process generating module 73, and the task obtaining module 71 is configured to obtain a task description text of a target task; the task disassembling module 72 is configured to obtain a plurality of subtasks constituting the target task based on the task description text; the flow generation module 73 is configured to, for each subtask: generating an implementation flow of the subtask, operating the implementation flow of the subtask to obtain an operation result, responding to the success of the implementation flow operation, checking the operation result to obtain whether the characterization operation result accords with an expected check result, obtaining a determination result for characterizing whether the implementation flow of the subtask is successful or not based on the check result, and selecting whether to continue to execute the step of generating the implementation flow of the subtask for the next subtask based on the determination result until the implementation flow of each subtask is successfully generated to obtain the implementation flow for automatically executing the target task.

In the above-mentioned scheme, the automatic flow generating device 70 obtains the task description text of the target task, obtains a plurality of subtasks constituting the target task based on the task description text, generates the implementation flow of the subtasks, and runs the implementation flow of the subtasks to obtain the running result, when the implementation flow is run successfully, checks the running result to obtain the checking result indicating whether the running result accords with the expected checking result, obtains the determination result indicating whether the implementation flow of the subtasks is generated successfully based on the checking result, selects whether to continue to execute the steps of the implementation flow of the subtasks for the next subtask based on the determination result until the implementation flow of each subtask is successfully generated, so as to obtain the implementation flow for automatically executing the target task, therefore, the automatic generation of the flow for implementing the target task can be triggered based on the task description text about the target task, and the generation quality of the implementation flow is judged based on the running result and the determination result, so that the generation efficiency and quality of the automatic flow can be improved.

In some disclosed embodiments, the task disassembly module 72 further includes a function selection module (not shown) for selecting a function matching the subtask as an objective function based on the description document of each function in the flow library; and calling the objective function to generate an implementation flow of the subtasks.

In some disclosed embodiments, the automated process generating device 70 further includes an error reporting operation module, at least for, in response to the implementation process operation failure, regenerating the implementation process of the subtask based on error reporting information of the implementation process operation failure of the subtask in the operation result; or in response to the implementation flow operation failure, correcting the description document of the function in the flow library based on the error reporting information of the implementation flow operation failure of the subtask in the operation result; the realization flow of the subtasks is generated by selecting and calling the function functions based on the description documents of the function functions in the flow library.

In some disclosed embodiments, the flow generation module 73 further includes a step determination module (not shown) for detecting a success or failure of the determination of the generation of the characterization of the result; responding to the determined result to represent successful generation, and continuously executing the steps of the realization flow of the generation subtask for the next subtask; and responding to the failure of determining the result representation generation, acquiring an analysis result of whether the target task is reasonably decomposed into a plurality of subtasks, and returning to a text based on task description based on the analysis result to acquire a plurality of subtasks constituting the target task.

In some disclosed embodiments, the step determining module further includes an analysis determining module (not shown) configured to, in response to the analysis result, characterize that the target task is reasonably decomposed into a plurality of subtasks, obtain an opinion description text of the target object decomposing the target task, and return to the step of obtaining a plurality of subtasks constituting the target task based on the task description text in combination with the opinion description text; and responding to the analysis result to represent that the target task is not reasonably decomposed into a plurality of subtasks, and returning to the step of obtaining a plurality of subtasks forming the target task based on the task description text.

In some disclosed embodiments, the task description text describes the target task in natural language, and the flow generation module 73 further includes a determination result acquisition module (not shown) for acquiring a determination result that the implementation flow of the characterization subtask is successful in response to the verification result characterizing that the operation result meets the expectation; and responding to the verification result to represent that the operation result does not accord with the expectation, and acquiring a determination result representing that the realization flow of the subtask fails to be generated.

Referring to fig. 8, fig. 8 is a schematic diagram of a frame of an embodiment of an automatic process generating apparatus 80 according to the present application, where the automatic process generating apparatus 80 includes a text obtaining module 81, a task decomposing module 82, a first sending module 83, a first receiving module 84, and a step selecting module 85, and the text obtaining module 81 is configured to obtain a task description text of a target task; the task decomposition module 82 is configured to obtain a plurality of subtasks constituting a target task based on the task description text; the first sending module 83 is configured to send the subtasks to the flow generating agent; the process generation agent generates an implementation process of a subtask, and sends the implementation process of the subtask to the process execution agent, the implementation process of the process execution agent operates the subtask to obtain an operation result, and responds to the successful operation of the implementation process, sends the operation result to the execution check agent, executes the check agent to check the operation result, obtains whether the characterization operation result accords with an expected check result, and sends the check result to the cooperative hub agent; the first receiving module 84 is configured to receive a verification result from the execution verification agent, and obtain a determination result indicating whether the implementation flow of the subtask is successful based on the verification result; the step selection module 85 is configured to select, based on the determination result, whether to return the step of sending the subtasks to the flow generation agent until the implementation flow of each subtask is successfully generated, so as to obtain the implementation flow for automatically executing the target task.

Therefore, the cooperative hub agent in the automatic flow generating device 80 obtains the task description text of the target task, obtains a plurality of subtasks constituting the target task based on the task description text, and sends the subtasks to the flow generating agent, the flow generating agent receives the subtasks from the cooperative hub agent, and generates an implementation flow of the subtasks, the implementation flow of the subtasks is sent to the flow executing agent, the flow executing agent receives the implementation flow of the subtasks from the flow generating agent, and then runs the implementation flow of the subtasks to obtain an operation result, and responds to the implementation flow successful, sends the operation result to the executing checking agent, executes the checking agent to receive the operation result from the flow executing agent, and then checks the operation result, obtains whether the operation result accords with the expected check result, sends the checking result to the cooperative hub agent, and then obtains a determination result for whether the implementation flow featuring the subtasks generates success based on the check result, and selects whether to return the step of the flow generating agent based on the determination result until the implementation flow of each subtask is successfully generated, thus the interactive text of the target task can be automatically generated based on the realization efficiency of the realization of the target task, and the interactive text can be automatically generated.

In some disclosed embodiments, the cooperative hub agent includes a first large model, and the task decomposition module 82 further includes a subtask acquisition module (not shown) for performing task decomposition on the task description text based on the first large model to obtain a plurality of subtasks constituting the target task; wherein, the first large model performs parameter fine adjustment in advance based on human feedback reinforcement learning.

In some disclosed embodiments, the step selection module 85 further includes a return selection module (not shown) for detecting that the determination result characterizes the generation success or the generation failure; responding to the determined result to represent successful generation, and returning a subtask to the step of generating the intelligent agent in the process; and responding to the failure of determining the result representation generation, acquiring an analysis result of whether the target task is reasonably decomposed into a plurality of subtasks, and returning to a text based on task description based on the analysis result to acquire a plurality of subtasks constituting the target task.

Referring to fig. 9, fig. 9 is a schematic diagram of a framework of an embodiment of an automatic process generating apparatus 90 according to the present application, where the automatic process generating apparatus 90 includes a second receiving module 91 and a second sending module 92, and the second receiving module 91 is configured to receive a subtask from a cooperative hub agent; the cooperative hub agent acquires and analyzes a task description text of a target task to obtain a plurality of subtasks forming the target task, and sends the subtasks to the flow generation agent; the second sending module 92 is configured to generate an implementation flow of the subtask, and send the implementation flow of the subtask to the flow execution agent; the method comprises the steps that a process execution agent runs an implementation process of a subtask to obtain a running result, the implementation process is responded to the success of the implementation process, the running result is sent to an execution check agent, the check agent is executed to check the running result to obtain whether the characterization running result accords with an expected check result, the check result is sent to a cooperative hub agent, the cooperative hub agent obtains a determination result whether the implementation process of the characterization subtask is successful or not based on the check result, and the step of sending the subtask to the process generation agent is selected based on the determination result until the implementation process of each subtask is successfully generated, so that the implementation process for automatically executing the target task is obtained.

Therefore, the cooperative hub agent in the automatic flow generating device 90 obtains the task description text of the target task, obtains a plurality of subtasks constituting the target task based on the task description text, and sends the subtasks to the flow generating agent, the flow generating agent receives the subtasks from the cooperative hub agent, and generates an implementation flow of the subtasks, sends the implementation flow of the subtasks to the flow executing agent, and after the flow executing agent receives the implementation flow of the subtasks from the flow generating agent, the implementation flow of the operation subtasks obtains an operation result, and responds to the implementation flow successful operation, sends the operation result to the execution checking agent, executes the checking agent to receive the operation result from the flow executing agent, checks the operation result, obtains whether the operation result accords with the expected check result, sends the checking result to the cooperative hub agent, and after the flow generating agent receives the check result from the execution checking agent, obtains a determination result for whether the implementation flow featuring the subtasks generates success or not based on the check result, and selects whether to return the step of the flow generating agent based on the determination result until the implementation flow of each subtask is successfully generated, thus the interactive text related to the target task can be automatically generated, and the quality of the interactive description of the target task can be automatically improved.

In some disclosed embodiments, the automated process generating device 90 further includes a second error reporting module (not shown) configured to receive an operation result sent by the process execution agent in response to the implementation process operation failure after sending the implementation process of the subtask to the process execution agent; the operation result comprises error reporting information of failure in operation of the realization flow of the subtask; and returning to the step of generating the realization flow of the subtask by combining the error reporting information.

In some disclosed embodiments, the second sending module 92 further includes a second generating module (not illustrated) configured to select, as the objective function, a function that matches the subtask based on the description document of each function in the flow library; and calling the objective function to generate an implementation flow of the subtasks.

In some disclosed embodiments, the process generation agent includes a second large model, and the second generation module further includes a description document generation module (not shown) for generating a description document of the obtained function based on the implementation process of the second large model processing the function before selecting the function matching with the subtask as the objective function based on the description document of each function in the process library.

Referring to fig. 10, fig. 10 is a schematic diagram illustrating a framework of an embodiment of an automatic process generating apparatus 100 according to the present application, where the automatic process generating apparatus 100 includes a third receiving module 101, a code running module 102, and a third sending module 103, and the third receiving module 101 receives an implementation process of a subtask from a process generating agent; the method comprises the steps that a cooperative hub agent obtains and analyzes a task description text of a target task to obtain a plurality of subtasks forming the target task, the subtasks are sent to a flow generation agent, the flow generation agent generates an implementation flow of the subtasks, and the implementation flow of the subtasks is sent to a flow execution agent; the code operation module 102 operates the realization flow of the subtasks to obtain an operation result; the third sending module 103 sends an operation result to the execution verification agent in response to successful implementation of the process operation; and executing the verification operation result of the verification agent to obtain whether the characterization operation result accords with the expected verification result, sending the verification result to the cooperative hub agent, obtaining a determination result indicating whether the realization flow of the subtask is generated successfully based on the verification result by the cooperative hub agent, and selecting whether to return to the step of sending the subtask to the flow generating agent based on the determination result until the realization flow of each subtask is generated successfully so as to obtain the realization flow for automatically executing the target task.

Therefore, the cooperative hub agent in the automatic flow generating device 100 obtains the task description text of the target task, obtains a plurality of subtasks constituting the target task based on the task description text, and sends the subtasks to the flow generating agent, the flow generating agent receives the subtasks from the cooperative hub agent, and generates an implementation flow of the subtasks, sends the implementation flow of the subtasks to the flow executing agent, and after the flow executing agent receives the implementation flow of the subtasks from the flow generating agent, the implementation flow of the operation subtasks obtains an operation result, and responds to the implementation flow successful operation, sends the operation result to the execution checking agent, executes the checking agent to receive the operation result from the flow executing agent, checks the operation result, obtains whether the operation result accords with the expected check result, sends the checking result to the cooperative hub agent, and after the flow generating agent receives the check result from the execution checking agent, obtains a determination result for whether the implementation flow featuring the subtasks generates success or not based on the check result, and selects whether to return the step of the flow generating agent based on the determination result until the implementation flow of each subtask is successfully generated, thus the interactive text related to the target task can be automatically generated, and the quality of the interactive description of the target task can be automatically improved.

In some disclosed embodiments, the automated process generation apparatus 100 further includes a second error reporting module configured to send an operation result to the process generation agent in response to implementing a process operation failure; the operation result contains error reporting information of failure in operation of the sub-task realization flow, and the flow generation agent combines the error reporting information to regenerate the sub-task realization flow.

In some disclosed embodiments, the automated process generating apparatus 100 further includes a third error reporting module configured to send an operation result to the document correction agent in response to implementing a process operation failure; the operation result contains error reporting information of failure in operation of the sub-task realization flow, the sub-task realization flow is generated by a flow generating agent selecting and calling a function based on a description document of the function in a flow library, and a document correction agent corrects the description document based on the error reporting information.

Referring to fig. 11, fig. 11 is a schematic diagram illustrating a framework of an embodiment of an automatic flow generating apparatus 110 according to the present application, where the automatic flow generating apparatus 110 includes a fourth receiving module 111 and a fourth sending module 112, and the fourth receiving module 111 is configured to receive an operation result from a flow execution agent; the method comprises the steps that a cooperative hub agent obtains and analyzes a task description text of a target task to obtain a plurality of subtasks forming the target task, the subtasks are sent to a flow generation agent, the flow generation agent generates an implementation flow of the subtasks, the implementation flow of the subtasks is sent to a flow execution agent, the implementation flow of the flow execution agent operates the subtasks to obtain an operation result, and the operation result is sent to an execution verification agent in response to the success of the implementation flow operation; the fourth sending module 112 is configured to check the operation result, obtain whether the operation result meets an expected check result, and send the check result to the cooperative hub agent; the cooperative hub agent obtains a determination result indicating whether the realization flow of the subtask is successful or not based on the verification result, and selects whether to return to the step of sending the subtask to the flow generation agent or not based on the determination result until the realization flow of each subtask is successfully generated, so as to obtain the realization flow for automatically executing the target task.

Therefore, the cooperative hub agent in the automatic flow generating device 110 obtains the task description text of the target task, obtains a plurality of subtasks constituting the target task based on the task description text, and sends the subtasks to the flow generating agent, the flow generating agent receives the subtasks from the cooperative hub agent, and generates an implementation flow of the subtasks, sends the implementation flow of the subtasks to the flow executing agent, and after the flow executing agent receives the implementation flow of the subtasks from the flow generating agent, the implementation flow of the operation subtasks obtains an operation result, and responds to the implementation flow successful operation, sends the operation result to the execution checking agent, executes the checking agent to receive the operation result from the flow executing agent, checks the operation result, obtains whether the operation result accords with the expected check result, sends the checking result to the cooperative hub agent, and after the flow generating agent receives the check result from the execution checking agent, obtains a determination result for whether the implementation flow featuring the subtasks generates success or not based on the check result, and selects whether to return the step of the flow generating agent based on the determination result until the implementation flow of each subtask is successfully generated, thus the interactive text of the target task can be automatically generated based on the realization efficiency of the realization of the target task, and the interactive text can be automatically generated.

Referring to fig. 12, fig. 12 is a schematic diagram of a frame of an embodiment of an automatic process generating apparatus 120 according to the present application, where the automatic process generating apparatus 120 includes a task parsing module 121, a process generating module 122, a process executing module 123, a process checking module 124, and a cycle generating module 125, and a cooperative hub agent in the task parsing module 121 obtains a task description text of a target task, obtains a plurality of subtasks constituting the target task based on the task description text, and sends the subtasks to the process generating agent; the process generation agent in the process generation module 122 receives subtasks from the cooperative hub agent, generates an implementation process of the subtasks, and sends the implementation process of the subtasks to the process execution agent; the flow execution agent in the flow execution module 123 receives the implementation flow of the subtasks from the flow generation agent, runs the implementation flow of the subtasks to obtain a running result, and sends the running result to the execution verification agent in response to successful running of the implementation flow; the execution and verification agent in the flow verification module 124 receives the operation result from the flow execution agent, verifies the operation result to obtain a verification result indicating whether the operation result meets the expectation, and sends the verification result to the cooperative hub agent; the cooperative hub agent in the loop generation module 125 receives the verification result from the execution verification agent, obtains a determination result indicating whether the implementation process of the subtask is successful or not based on the verification result, and selects whether to return the step of sending the subtask to the process generation agent or not based on the determination result until the implementation process of each subtask is successfully generated, so as to obtain the implementation process for automatically executing the target task.

Therefore, the cooperative hub agent in the automatic flow generating device 120 obtains the task description text of the target task, obtains a plurality of subtasks constituting the target task based on the task description text, and sends the subtasks to the flow generating agent, the flow generating agent receives the subtasks from the cooperative hub agent, and generates an implementation flow of the subtasks, the implementation flow of the subtasks is sent to the flow executing agent, the flow executing agent receives the implementation flow of the subtasks from the flow generating agent, and then runs the implementation flow of the subtasks to obtain an operation result, and responds to the implementation flow successful, sends the operation result to the executing checking agent, executes the checking agent to receive the operation result from the flow executing agent, and then checks the operation result, obtains whether the operation result accords with the expected check result, sends the checking result to the cooperative hub agent, and then obtains a determination result for whether the implementation flow featuring the subtasks generates success based on the check result, and selects whether to return the step of the flow generating agent based on the determination result until the implementation flow of each subtask is successfully generated, thus the interactive text of the target task can be automatically generated based on the realization efficiency of the realization of the target task, and the interactive text can be automatically generated.

Referring to fig. 13, fig. 13 is a schematic diagram of a frame of an electronic device 130 according to an embodiment of the application. The electronic device 130 comprises a memory 131 and a processor 132, the memory 131 storing program instructions, the processor 132 being adapted to execute the program instructions to implement the steps of any of the above-described embodiments of the automated process generation method. Reference may be made specifically to the foregoing disclosed embodiments, and details are not repeated here. The electronic device 130 may include, in particular, but not limited to: servers, smartphones, notebook computers, tablet computers, etc., are not limited herein.

In particular, the processor 132 is configured to control itself and the memory 131 to implement the steps of any of the automated process generation method embodiments described above. The processor 132 may also be referred to as a CPU (Central Processing Unit ). The processor 132 may be an integrated circuit chip having signal processing capabilities. The Processor 132 may also be a general purpose Processor, a digital signal Processor (DIGITAL SIGNAL Processor, DSP), an Application SPECIFIC INTEGRATED Circuit (ASIC), a Field-Programmable gate array (Field-Programmable GATE ARRAY, FPGA) or other Programmable logic device, a discrete gate or transistor logic device, a discrete hardware component. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. In addition, the processor 132 may be commonly implemented by an integrated circuit chip.

In the above solution, the electronic device 130 obtains the task description text of the target task, obtains a plurality of subtasks constituting the target task based on the task description text, generates the implementation flow of the subtasks, and runs the implementation flow of the subtasks to obtain the running result, when the implementation flow runs successfully, checks the running result to obtain the check result indicating whether the running result meets the expectations, obtains the determination result indicating whether the implementation flow of the subtasks is successful based on the check result, and selects whether to continue to execute the steps of the implementation flow of the subtasks for the next subtask based on the determination result until the implementation flow of each subtask is successfully generated, so as to obtain the implementation flow for automatically executing the target task, thus triggering the automatic generation of the flow of the implementation target task based on the task description text about the target task, and judging the generation quality of the implementation flow based on the running result and the determination result, so that the generation efficiency and quality of the automatic flow can be improved.

Referring to fig. 14, fig. 14 is a schematic diagram illustrating an embodiment of an automated process generation system 140 according to the present application. The system comprises a cooperative hub agent 141, a flow generation agent 142, a flow execution agent 143 and an execution check agent 144 which are in communication connection with each other, wherein the cooperative hub agent 141 is used for executing the automated flow generation method in the above-disclosed embodiment, the flow generation agent 142 is used for executing the automated flow generation method in the above-disclosed embodiment, the flow execution agent 143 is used for executing the automated flow generation method in the above-disclosed embodiment, and the execution check agent 144 is used for executing the automated flow generation method in the above-disclosed embodiment.

In the above-mentioned scheme, the cooperative hub agent 141 in the automated process generation system 140 obtains the task description text of the target task, obtains a plurality of subtasks constituting the target task based on the task description text, and sends the subtasks to the process generation agent 142, the process generation agent 142 receives the subtasks from the cooperative hub agent 141 and generates the implementation process of the subtasks, sends the implementation process of the subtasks to the process execution agent 143, after the process execution agent 143 receives the implementation process of the subtasks from the process generation agent 142, the implementation process of the running subtasks obtains the running result, and responds to the successful implementation process, sends the running result to the execution verification agent 144, and checks the running result after the execution verification agent 144 receives the running result from the process execution agent 143, obtains the expected verification result representing whether the running result accords with the expected verification result, and sends the verification result to the cooperative hub agent 141, obtains the successful determination result representing whether the implementation process of the subtasks is generated based on the verification result, and selects whether to send the subtasks back to the step of the generation agent 142 based on the determination result, thus the quality of the realization of the subtasks can be automatically triggered based on the realization of the realization process of the target task, and the quality of the interactive tasks can be automatically generated based on the quality of the realization of the target task.

Referring to FIG. 15, FIG. 15 is a schematic diagram of a computer readable storage medium 150 according to an embodiment of the application. The computer readable storage medium 150 stores program instructions 151 that can be executed by a processor, the program instructions 151 for implementing the steps in any of the above-described embodiments of an automated process generation method.

In some embodiments, functions or modules included in an apparatus provided by the embodiments of the present disclosure may be used to perform a method described in the foregoing method embodiments, and specific implementations thereof may refer to descriptions of the foregoing method embodiments, which are not repeated herein for brevity.

The foregoing description of various embodiments is intended to highlight differences between the various embodiments, which may be the same or similar to each other by reference, and is not repeated herein for the sake of brevity.

In the several embodiments provided in the present application, it should be understood that the disclosed method and apparatus may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of modules or units is merely a logical functional division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical, or other forms.

The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed over a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiment.

In addition, each functional unit in the embodiments of the present application may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units.

The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be embodied in essence or a part contributing to the prior art or all or part of the technical solution in the form of a software product stored in a storage medium, including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) or a processor (processor) to execute all or part of the steps of the methods of the embodiments of the present application. And the aforementioned storage medium includes: a usb disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.

If the technical scheme of the application relates to personal information, the product applying the technical scheme of the application clearly informs the personal information processing rule before processing the personal information and obtains the autonomous agreement of the individual. If the technical scheme of the application relates to sensitive personal information, the product applying the technical scheme of the application obtains individual consent before processing the sensitive personal information, and simultaneously meets the requirement of 'explicit consent'. For example, a clear and remarkable mark is set at a personal information acquisition device such as a camera to inform that the personal information acquisition range is entered, personal information is acquired, and if the personal voluntarily enters the acquisition range, the personal information is considered as consent to be acquired; or on the device for processing the personal information, under the condition that obvious identification/information is utilized to inform the personal information processing rule, personal authorization is obtained by popup information or a person is requested to upload personal information and the like; the personal information processing rule may include information such as a personal information processor, a personal information processing purpose, a processing mode, and a type of personal information to be processed.

Claims

1. An automated process generation method, comprising:

acquiring a task description text of a target task;

based on the task description text, obtaining a plurality of subtasks forming the target task;

For each of the subtasks: generating an implementation flow of the subtask, operating the implementation flow of the subtask to obtain an operation result, responding to the successful operation of the implementation flow, checking the operation result to obtain a check result which indicates whether the operation result accords with expectations, obtaining a determination result which indicates whether the implementation flow of the subtask is successful or not based on the check result, and selecting whether to continue executing the steps for generating the implementation flow of the subtask on the next subtask based on the determination result until the implementation flow of each subtask is successfully generated so as to obtain the implementation flow for automatically executing the target task;

the step of selecting whether to continue to execute the realization flow for generating the subtask for the next subtask based on the determination result includes:

Detecting success or failure of the generation of the determined result characterization;

responding to the determined result to represent successful generation, and continuously executing the steps of the realization flow for generating the subtasks for the next subtask;

And responding to the failure of determining result representation generation, acquiring an analysis result for representing whether the target task is reasonably decomposed into a plurality of subtasks, and returning the text based on task description based on the analysis result to obtain a plurality of subtasks forming the target task.

2. The method of claim 1, wherein generating the implementation flow of the subtask comprises:

selecting a function matched with the subtask as an objective function based on the explanation document of each function in the flow library;

and calling the objective function to generate the realization flow of the subtask.

3. The method of claim 1, further comprising at least one of:

Responding to the operation failure of the realization flow, and regenerating the realization flow of the subtask based on the error reporting information of the operation failure of the realization flow of the subtask in the operation result;

Responding to the operation failure of the realization flow, and correcting the description document of the function in the flow library based on the error reporting information of the operation failure of the realization flow of the subtasks in the operation result; the realization flow of the subtasks is generated by selecting and calling the function functions based on the description documents of the function functions in the flow library.

4. The method according to claim 1, wherein the step of returning the task description text based on the analysis result to obtain a number of sub-tasks constituting the target task comprises at least one of:

responding to the analysis result to represent that the target task is reasonably decomposed into a plurality of subtasks, obtaining an opinion description text of decomposing the target task by a target object, and returning the task description text to obtain a plurality of subtasks forming the target task;

and responding to the analysis result to represent that the target task is not reasonably decomposed into a plurality of subtasks, and returning the text based on the task description to obtain a plurality of subtasks constituting the target task.

5. The method according to any one of claims 1 to 4, wherein the task description text describes the target task in natural language;

and/or, based on the verification result, obtaining a determination result indicating whether the implementation flow of the subtask generates success, including at least one of the following: responding to the verification result to represent that the operation result accords with the expectation, and acquiring a determination result representing that the realization flow of the subtask is successfully generated; and responding to the verification result to represent that the operation result does not accord with the expectation, and acquiring a determination result representing that the realization flow of the subtask fails to be generated.

6. An automated process generation method, comprising:

acquiring a task description text of a target task;

Sending the subtasks to a flow generating agent; the process generating agent generates an implementation process of the subtask and sends the implementation process of the subtask to a process execution agent, the process execution agent runs the implementation process of the subtask to obtain a running result, the running result is sent to an execution checking agent in response to successful running of the implementation process, the execution checking agent checks the running result to obtain a checking result which characterizes whether the running result accords with expectations, and the checking result is sent to a cooperative hub agent;

Receiving a verification result from the execution verification agent, and obtaining a determination result indicating whether the realization flow of the subtask is successful or not based on the verification result;

selecting whether to return the step of sending the subtasks to a flow generation agent or not based on the determination result until the realization flow of each subtask is successfully generated, so as to obtain the realization flow for automatically executing the target task;

The step of selecting whether to return the subtask to the flow generating agent based on the determination result includes:

responding to the determined result to represent successful generation, and returning the step of sending the subtasks to a flow generating agent;

7. The method of claim 6, wherein the collaborative hub agent comprises a first large model, wherein the deriving sub-tasks that comprise the target task based on the task description text comprises:

performing task decomposition on the task description text based on the first large model to obtain a plurality of subtasks forming the target task;

Wherein the first large model performs parameter fine adjustment in advance based on human feedback reinforcement learning.

8. An automated process generation method, comprising:

receiving subtasks from the cooperative hub agent; the cooperative hub agent acquires and analyzes a task description text of a target task to obtain a plurality of subtasks forming the target task, and sends the subtasks to a flow generation agent;

Generating an implementation flow of the subtasks, and sending the implementation flow of the subtasks to a flow execution agent; the step of obtaining a determination result indicating whether the realization process of the subtask is successful or not by the cooperative pivot agent based on the verification result, and selecting whether to return the step of sending the subtask to the process generation agent or not based on the determination result until the realization process of each subtask is successfully generated, so as to obtain the realization process for automatically executing the target task, and selecting whether to return the step of sending the subtask to the process generation agent based on the determination result, wherein the step of obtaining the realization process for automatically executing the target task comprises the steps of:

and responding to the determining result to represent the generation failure, acquiring an analysis result of whether the target task is reasonably decomposed into a plurality of subtasks, and returning a task description text of the analysis target task based on the analysis result to acquire a plurality of subtasks forming the target task.

9. The method of claim 8, wherein after said sending the execution flow of the subtask to a flow execution agent, the method further comprises:

Receiving an operation result sent by the flow execution agent in response to the failure of implementing the flow operation; the operation result comprises error reporting information of failure in operation of the realization flow of the subtask;

And returning the step of generating the realization flow of the subtask by combining the error reporting information.

10. The method of claim 8, wherein the flow of the realization of the subtasks is generated by the flow generating agent selecting and invoking a function of each function in a flow library based on a description document of the function, the flow generating agent containing a second bigram, the method further comprising:

And processing the implementation flow of the function based on the second large model, and generating a description document for obtaining the function.

11. An automated process generation method, comprising:

Receiving an implementation flow of a subtask from a flow generation agent; the method comprises the steps that a cooperative hub agent acquires and analyzes a task description text of a target task to obtain a plurality of subtasks forming the target task, the subtasks are sent to a flow generation agent, the flow generation agent generates an implementation flow of the subtasks, and the implementation flow of the subtasks is sent to a flow execution agent;

operating the realization flow of the subtasks to obtain an operation result;

responding to successful operation of the implementation flow, and sending the operation result to an execution verification agent; the step of executing the verification agent to verify the operation result, obtaining a verification result representing whether the operation result meets expectations, and sending the verification result to the cooperative hub agent, wherein the cooperative hub agent obtains a determination result representing whether the realization process of the subtasks generates success based on the verification result, selects whether to return the step of sending the subtasks to the process generation agent based on the determination result until the realization process of each subtask is successfully generated, so as to obtain a realization process for automatically executing the target task, and the step of selecting whether to return the transmission of the subtasks to the process generation agent based on the determination result comprises the following steps:

12. The method of claim 11, wherein the method further comprises:

Responding to the failure of implementing the process operation, and sending the operation result to the process generation agent; the operation result contains error reporting information of failure in operation of the sub-task realization flow, and the flow generation agent is combined with the error reporting information to regenerate the sub-task realization flow.

13. The method of claim 11, wherein the method further comprises:

Responding to the failure of the operation of the realization flow, and sending the operation result to a document correction agent; the operation result comprises error reporting information of failure in operation of the sub-task realization flow, the sub-task realization flow is generated by the flow generation agent selecting and calling a function based on a description document of the function in a flow library, and the document correction agent corrects the description document based on the error reporting information.

14. An automated process generation method, comprising:

receiving an operation result from a flow execution agent; the method comprises the steps that a cooperative hub agent obtains and analyzes a task description text of a target task to obtain a plurality of subtasks forming the target task, the subtasks are sent to a flow generation agent, the flow generation agent generates an implementation flow of the subtasks and sends the implementation flow of the subtasks to a flow execution agent, the flow execution agent runs the implementation flow of the subtasks to obtain the running result, and the running result is sent to an execution verification agent in response to successful running of the implementation flow;

Checking the operation result to obtain a check result which characterizes whether the operation result accords with the expectation, and sending the check result to the cooperative hub intelligent agent; the step of selecting whether to return the sub-task to the process generation agent based on the determination result, until the realization process of each sub-task is successfully generated, so as to obtain the realization process for automatically executing the target task, and the step of selecting whether to return the sub-task to the process generation agent based on the determination result, wherein the step of selecting whether to return the sub-task to the process generation agent comprises the following steps:

15. An automated process generation method, comprising:

The cooperative pivot agent acquires a task description text of a target task, obtains a plurality of subtasks forming the target task based on the task description text, and sends the subtasks to a flow generation agent;

The process generating agent receives the subtasks from the cooperative hub agent, generates an implementation process of the subtasks, and sends the implementation process of the subtasks to the process executing agent;

The flow execution agent receives the realization flow of the subtasks from the flow generation agent, operates the realization flow of the subtasks to obtain an operation result, and responds to the successful operation of the realization flow, and sends the operation result to the execution verification agent;

The execution verification agent receives an operation result from the process execution agent, verifies the operation result to obtain a verification result which characterizes whether the operation result accords with an expected or not, and sends the verification result to the cooperative hub agent;

The cooperative hub agent receives a verification result from the execution verification agent, obtains a determination result representing whether the realization flow of the subtask is generated successfully or not based on the verification result, and selects whether to return the step of sending the subtask to the flow generation agent or not based on the determination result until the realization flow of each subtask is generated successfully, so as to obtain the realization flow for automatically executing the target task;

16. An electronic device comprising a memory and a processor coupled to each other, the memory having program instructions stored therein, the processor configured to execute the program instructions to implement the automated process generation method of any one of claims 1-15.

17. An automated process generation system comprising a cooperative hub agent, a process generation agent, a process execution agent, and an execution verification agent in communicative connection with each other, and wherein the cooperative hub agent is configured to perform the automated process generation method of any one of claims 6 to 7, the process generation agent is configured to perform the automated process generation method of any one of claims 8 to 10, the process execution agent is configured to perform the automated process generation method of any one of claims 11 to 13, and the execution verification agent is configured to perform the automated process generation method of claim 14.

18. A computer readable storage medium, characterized in that program instructions executable by a processor for implementing the automated process generation method of any one of claims 1 to 15 are stored.