Cross-system business fusion workflow generation method and system based on large model

Abstract

The invention discloses a cross-system business fusion workflow generation method and system based on a large model, and the method comprises the steps: obtaining a task demand, decomposing the task demand through a preset large model, and obtaining a plurality of subtask description steps; retrieving the sub-task description steps to obtain a plurality of pieces of related component information, and inputting the task demand, the sub-task description steps and the related component information into the large model to obtain component description information; and constructing a prompt template, and obtaining a workflow through the large model according to the prompt template, the task demand and the component description information. The working efficiency of business flow generation can be greatly improved, the labor cost can be reduced, the accuracy and applicability of workflow generation can be improved, and the method can be widely applied to the technical field of data processing.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method and system for generating a cross-system business fusion workflow based on a large model. Background Art

[0002] As digital transformation accelerates, enterprises increasingly rely on automated workflows to improve operational efficiency and responsiveness. Traditional manual management and custom development methods can no longer meet the growing needs for flexibility and integration. Faced with the challenges of cross-system business process management, the industry has gradually turned to using advanced computer science methods and technologies to simplify and automate these processes. In recent years, artificial intelligence technology, especially the development of large language models, has provided powerful tools for the creation of automated workflows. These models, due to their excellent language understanding and generation capabilities, can handle complex instructions and task descriptions, bringing new possibilities for achieving automated workflow generation.

[0003] Although large language models have shown great potential in the field of automated workflow generation, they still face some key challenges. On the one hand, existing technical solutions often experience error accumulation when generating multi-step workflows, especially when dealing with long sequence tasks. The model may produce logical errors, fictional plots, or content that is inconsistent with the facts. On the other hand, existing workflow generation methods usually lack an in-depth understanding of users' actual needs, resulting in the generated workflows not meeting expectations or being difficult to directly apply to actual business scenarios. In addition, existing workflow generation methods rarely consider the problem of cross-system component integration, which makes it difficult for workflows to run seamlessly in different IT environments. These problems limit the actual application scope and efficiency of automated workflow technology. Summary of the invention

[0004] In order to solve the above technical problems, the purpose of the present invention is to provide a method and system for generating a cross-system business fusion workflow based on a large model with high accuracy and applicability.

[0005] To achieve the above purpose, an embodiment of the present application proposes a method for generating a cross-system business integration workflow based on a large model, comprising the following steps:

[0006] Obtaining task requirements, decomposing the task requirements through a preset large model, and obtaining multiple subtask description steps;

[0007] Retrieving each of the subtask description steps to obtain a plurality of related component information, and then inputting the task requirements, the subtask description steps and the related component information into the large model to obtain component description information;

[0008] A prompt template is constructed, and a workflow is obtained according to the prompt template, the task requirements and the component description information through the large model.

[0009] In some embodiments, the task requirement is decomposed by a preset large model to obtain multiple subtask description steps, specifically including:

[0010] Determine the optimal strategy of the large model and input the task requirements into the large model;

[0011] Obtaining multiple states according to the task requirements, and obtaining cumulative benefits and heuristic values ​​according to all the states;

[0012] Calculating the expansion value corresponding to each of the states according to the accumulated income and the heuristic value;

[0013] Sort the states from high to low according to the extension value, and determine a preset number of the states as candidate states according to the sorting result;

[0014] Each of the candidate states is expanded according to the optimal strategy to obtain a plurality of subtask description steps.

[0015] In some embodiments, the multiple states are obtained according to the task requirements, specifically:

[0016] The decomposition process of the task requirements is conceptualized as a Markov decision framework, which includes a plurality of states, a plurality of actions, a discount factor, and a reward function for a state-action pair.

[0017] In some embodiments, the cumulative benefits and heuristic values ​​are obtained according to all the states, specifically including:

[0018] Calculate the rewards of all the states, perform aggregation operations on the rewards, and obtain the cumulative income;

[0019] The optimal cumulative expected reward value of the state-action pair is calculated according to the reward function and the discount factor, and the optimal cumulative expected reward value is used as the heuristic value.

[0020] In some embodiments, the step of retrieving each of the subtask descriptions to obtain a plurality of related component information specifically includes:

[0021] Obtain a component library, wherein the component library includes multiple components;

[0022] Extracting a semantically-aware embedding vector and a query embedding vector of each of the components in the component library;

[0023] Obtaining the relevance of the subtask description step and the component according to the semantically-aware embedding vector and the query embedding vector;

[0024] The component with the highest correlation corresponding to each of the subtask description steps is determined to obtain a plurality of related component information.

[0025] In some embodiments, the construction prompt template is specifically:

[0026] Constructing the prompt template according to the component description information and the subtask description steps;

[0027] The prompt template includes role and target prompts, context information prompts, component description prompts, restriction prompts and demonstration prompts.

[0028] In some embodiments, obtaining the workflow according to the prompt template, the task requirements and the component description information through the large model specifically includes:

[0029] Input the prompt template and the task requirement into the large model to obtain a workflow description file;

[0030] A user feedback result is obtained according to the workflow description file, and the workflow description file is updated according to the user feedback result through the large model to obtain the workflow.

[0031] To achieve the above purpose, another aspect of the embodiment of the present application proposes a cross-system business integration workflow generation system based on a large model, including:

[0032] A task decomposition module is used to obtain task requirements and decompose the task requirements through a large language model to obtain multiple subtask description steps;

[0033] A component generation module, used to retrieve each of the subtask description steps to obtain a plurality of related component information, and then input the task requirements, the subtask description steps and the related component information into a preset macro model to obtain component description information;

[0034] The workflow generation module is used to construct a prompt template, and obtain the workflow according to the prompt template, the task requirements and the component description information through the large model.

[0035] To achieve the above-mentioned purpose, another aspect of an embodiment of the present application proposes an electronic device, which includes a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for realizing connection and communication between the processor and the memory. When the program is executed by the processor, the cross-system business fusion workflow generation method based on a large model as described above is realized.

[0036] To achieve the above-mentioned purpose, another aspect of an embodiment of the present application proposes a storage medium, which is a computer-readable storage medium used for computer-readable storage, and the storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the cross-system business fusion workflow generation method based on a large model as described above.

[0037] The beneficial effects of the present invention are as follows: the cross-system business fusion workflow generation method and system based on the big model of the present invention first obtains the task requirements, decomposes the task requirements through the preset big model, obtains multiple subtask description steps, then retrieves each subtask description step, obtains multiple related component information, and then inputs the task requirements, subtask description steps and related component information into the big model to obtain component description information, and finally constructs a prompt template, and obtains the workflow according to the prompt template, task requirements and component description information through the big model. The present invention automatically decomposes the task requirements described by the user into a series of operable subtask description steps through the big model, and automatically retrieves the most relevant component description information according to the subtask description steps, and finally automatically generates a structured and executable workflow, which can greatly improve the work efficiency of business flow generation, reduce labor costs, and improve the accuracy and applicability of workflow generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solution in the embodiments of the present invention, the following introduction is made to the drawings required for use in the embodiments of the present invention. It should be understood that the drawings introduced below are only for the convenience of clearly describing some embodiments of the technical solution of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0039] Figure 1 A flowchart of a method for generating a cross-system business fusion workflow based on a large model provided by an embodiment of the present invention;

[0040] Figure 2 A schematic diagram of a flow chart of assembly of components provided by an embodiment of the present invention;

[0041] Figure 3A schematic diagram of the structure of a cross-system business fusion workflow generation system based on a large model provided in an embodiment of the present invention;

[0042] Figure 4 A schematic diagram of the hardware structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is further described in detail below in conjunction with the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of the present application. They are only examples of devices and methods consistent with some aspects of the embodiments of the present application as detailed in the attached claims.

[0044] It is understood that the terms "first", "second", etc. used in this application can be used to describe various concepts in this article, but unless otherwise specified, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another concept. For example, without departing from the scope of the embodiment of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the words "if" and "if" as used herein can be interpreted as "at the time of" or "when" or "in response to determination".

[0045] The terms "at least one", "multiple", "each", "any", etc. used in this application, at least one includes one, two or more, multiple includes two or more, each refers to each of the corresponding multiple, and any refers to any one of the multiple.

[0046] As digital transformation accelerates, enterprises increasingly rely on automated workflows to improve operational efficiency and responsiveness. Traditional manual management and custom development methods can no longer meet the growing needs for flexibility and integration. Faced with the challenges of cross-system business process management, the industry has gradually turned to using advanced computer science methods and technologies to simplify and automate these processes. In recent years, artificial intelligence technology, especially the development of large language models, has provided powerful tools for the creation of automated workflows. These models, due to their excellent language understanding and generation capabilities, can handle complex instructions and task descriptions, bringing new possibilities for achieving automated workflow generation.

[0047] Although large language models have shown great potential in the field of automated workflow generation, they still face some key challenges. On the one hand, existing technical solutions often experience error accumulation when generating multi-step workflows, especially when dealing with long sequence tasks. The model may produce logical errors, fictional plots, or content that is inconsistent with the facts. On the other hand, existing workflow generation methods usually lack an in-depth understanding of users' actual needs, resulting in the generated workflows not meeting expectations or being difficult to directly apply to actual business scenarios. In addition, existing workflow generation methods rarely consider the problem of cross-system component integration, which makes it difficult for workflows to run seamlessly in different IT environments. These problems limit the actual application scope and efficiency of automated workflow technology.

[0048] A method for automated workflow generation based on a rule engine and an expert system in the prior art relies on preset rules and expert knowledge to guide the creation of workflows. This method uses a rule engine to execute business rules and combines it with an expert system to provide domain expertise. The expert system is responsible for defining the rules, while the rule engine guides the generation and execution of workflows based on these rules. However, this method is less flexible because it runs based on a fixed set of rules. This means that when faced with changes in business requirements, the rules and expert knowledge base need to be manually updated, which increases maintenance costs and response time. In addition, this method has limited ability to handle unexpected situations, which may lead to interruptions or incorrect execution of workflows.

[0049] A widely adopted approach in the prior art is to optimize the performance of large models in specific application scenarios by performing domain-specific fine-tuning on them. This strategy uses language models as a basis and further fine-tunes them on datasets focused on a certain professional field. Through fine-tuning, the model can absorb and understand the unique terminology, conceptual framework, and language habits of the field, thereby demonstrating more accurate and professional output capabilities when handling related tasks. However, although the fine-tuning strategy can improve the model's capabilities in specific fields to a certain extent, it has several significant disadvantages. First, the fine-tuning process requires a large amount of well-annotated in-domain data, which is not only costly, but also for some niche or emerging fields, obtaining sufficient datasets is itself a major challenge. Secondly, fine-tuning may lead to a decrease in the generalization ability of the model, that is, poor performance when dealing with out-of-domain or edge cases.

[0050] To this end, an embodiment of the present invention proposes a cross-system business fusion workflow generation method based on a large model, first obtaining task requirements, decomposing the task requirements through a preset large model, obtaining multiple subtask description steps, then retrieving each subtask description step, obtaining multiple related component information, and then inputting the task requirements, subtask description steps and related component information into the large model to obtain component description information, and finally constructing a prompt template, and obtaining a workflow according to the prompt template, task requirements and component description information through the large model. The present invention automatically decomposes the task requirements described by the user into a series of operable subtask description steps through a large model, and automatically retrieves the most relevant component description information according to the subtask description steps, and finally automatically generates a structured and executable workflow, which can greatly improve the work efficiency of business flow generation, reduce labor costs, and improve the accuracy and applicability of workflow generation.

[0051] Reference Figure 1 , Figure 1 A flowchart of a method for generating a cross-system business integration workflow based on a large model is provided in an embodiment of the present invention. The embodiment of the present invention proposes a method for generating a cross-system business integration workflow based on a large model, and the method includes steps S101 to S103:

[0052] S101, obtaining task requirements, decomposing the task requirements through a preset large model, and obtaining multiple subtask description steps;

[0053] Specifically, the large model generates a series of subtask description steps from the task requirements described by the user to solve the task. These subtask description steps serve as an explanatory framework to decompose the original task requirements into specific and actionable criteria. Specifically, taking the task requirement q as input, the large model divides the solution steps into multiple subtask description steps with the goal of completing the task, where the final task sequence a can be regarded as the concatenation of these T single-step subtasks, formalized as a=a1; a1; ...a T Each subtask description step is one or more lines of tokens output by the large model.

[0054] As an optional implementation, the task requirements are decomposed by a preset large model to obtain multiple subtask description steps, which can be specifically divided into the following steps S1011 to S1015:

[0055] S1011. Determine the optimal strategy of the large model and input the task requirements into the large model;

[0056] Specifically, the solution generated in the embodiment of the present invention must match the ground-truth to be given a reward value of 1. θIt is embodied by a large model (such as LLMs), which generates a series of subtask description steps according to the task requirement q, and the optimal strategy π θ As shown below:

[0057]

[0058] S1012, obtaining multiple states according to task requirements, and obtaining cumulative benefits and heuristic values ​​according to all states;

[0059] As an optional implementation, the step of obtaining multiple states according to task requirements can be specifically divided into the following steps S10121:

[0060] S10121. Conceptualize the decomposition process of task requirements as a Markov decision framework, which includes multiple states, multiple actions, discount factors, and reward functions for state-action pairs.

[0061] Specifically, the task decomposition process is conceptualized as a Markov decision process (MDP) Among them, the status represents the concatenation of the subtask description steps generated up to step t-1 and the input task requirements (i.e., s t =[q;a1;…a t-1 ]), in the special case s1=q. Action Indicates that the large model is inputting the current state s t The next subtask description step is generated after. From the current state s t To the next state s t+1 Deterministic state transfer This is done by a simple concatenation operation. γ is the discount factor. It is a reward function that measures the degree of problem solving. The reward function is result-based. It rewards by comparing the final result with the ground-truth. The reward function of the state-action pair As shown below:

[0062]

[0063] As an optional implementation, the step of obtaining the cumulative benefits and heuristic values ​​according to all states can be specifically divided into the following steps S10122 and S10123:

[0064] S10122. Calculate the rewards of all states, perform aggregation operations on each reward, and obtain cumulative benefits;

[0065] Specifically, using a process-based reward function To calculate from the initial state s1 to the current state st The cumulative benefit g(s t ), as shown below:

[0066]

[0067] Among them, Agg∈{min,max,∑,[-1]},[-1] means taking the reward of the last state as the benefit, which is the summary from s1 to s t The aggregation function of the rewards on the paths .

[0068] S10123. Calculate the optimal cumulative expected return value of the state-action pair according to the reward function and the discount factor, and then use the optimal cumulative expected return value as the heuristic value.

[0069] Specifically, using the state-action pair (s t ,a t )’s optimal cumulative expected return value Q * (s t ,a t ) as the heuristic value h(s) t ).h(s t ) is the heuristic value of reaching the correct answer from the current state. It is an estimate that measures the likelihood or probability of reaching the final correct answer from the current state.

[0070] The optimal cumulative expected return value Q * (s t ,a t ) satisfies the Bellman optimal equation:

[0071]

[0072] S1013, calculating the expansion value corresponding to each state according to the cumulative income and the heuristic value;

[0073] It should be noted that large models use autoregressive methods to generate natural language, which means that they predict the next token in the sequence based on previously generated tokens. However, when generating subtask description steps, if any of the previous steps are wrong, the large model may introduce errors, fictitious plots, and contradictory statements in the subsequent reasoning process, thus failing to effectively solve the problem. Since large models generate each token with limited computing resources, they cannot invest more computing resources in solving difficult problems. Therefore, the embodiment of the present invention proposes Q * , that is, a * The algorithm's large model task decomposition algorithm solves this problem. This framework can effectively guide the large model to choose the most promising next step when performing multi-step reasoning.

[0074] Specifically, we consider finding a solution to the subtask description step of a given task description as a heuristic search process, in which each state s t are associated with an f value, which is used to estimate if the extended state s t How much expansion value will be obtained, expansion value f(s t ) is calculated by the following formula:

[0075] f(s t )=g(s t )+λh(s t )

[0076] Among them, g(s t ) represents the cumulative benefit from the initial state s1, h(s t ) represents the heuristic value, and λ is used to balance g(s t ) and h(s t ) is the weight coefficient between them.

[0077] The state-action pair (s t ,a t )’s optimal cumulative expected return value Q * (s t ,a t ) as the heuristic value h(s) t ), then the extended value f(s t ) can be transformed into the following formula:

[0078]

[0079] Since it is unrealistic to enumerate all possible next reasoning steps, the embodiment of the present invention limits the selection range to only the first K of all step candidates returned by the large model, so the above formula can be written as:

[0080]

[0081] Targeting Q * The most critical problem in the problem is to estimate the optimal Q value (i.e. the optimal cumulative expected return value) of the state-action pair for a given task description problem. This requires learning a proxy Q value model. From the dataset The approximate estimate Q * , the dataset D consists of N task requirements, each of which has a set of M trajectories These trajectories are sampled from the optimal policy.

[0082] Formally, the proxy Q-value model It is learned by minimizing the following objective function:

[0083]

[0084] Among them, s t It's a track The partial reasoning trajectory from step t-1 to step t-1, is the true optimal Q value Q * (s t ,a t ) is similar to the label .

[0085] Specifically, the embodiment of the present invention adopts an offline reinforcement learning method to construct a Q value label for the task requirements. For a given offline dataset D, the proxy Q-value model is learned using Fitted Q-iteration. For each iteration l, the Q-value label is constructed by the following formula:

[0086]

[0087] in, is the proxy Q value model learned in iteration l-1. After that, a new proxy model is trained according to the above objective function These two phases will be repeated L times, using As a Proxy Q-Value Model Make inferences.

[0088] In obtaining the proxy Q value model After that, insert it into the following formula to calculate the eigenvalue of each state and perform the * Best-first search:

[0089]

[0090] S1014, sorting the states from high to low according to the extended values, and determining a preset number of states as candidate states according to the sorting results;

[0091] S1015. Expand each candidate state according to the optimal strategy to obtain multiple subtask description steps.

[0092] Specifically, based on A *The algorithm's large model task decomposition algorithm controls the best-first search process by designing two sets: one set contains candidate states that need to be explored, called the unvisited set (unvisited), and the other set records the states that have been visited (visited). At each step, the algorithm selects the state with the largest f value from the unvisited set and expands the state by generating the top K (i.e., the preset number) alternative actions through the optimal strategy. Then, the algorithm updates the two sets and repeats this process until a terminal state is found, i.e., a complete reasoning trajectory. Finally, the answer is extracted from the terminal state as the final result, and the complete subtask description steps are obtained.

[0093] It should be noted that the preset number can be selected according to actual needs. For example, the preset number can be 5, 10, etc.

[0094] S102, searching for each subtask description step to obtain a plurality of related component information, and then inputting the task requirements, subtask description steps and related component information into the big model to obtain component description information;

[0095] It should be noted that, after obtaining the complete subtask description step a=a1;a1;…a T After that, it is necessary to generate corresponding component description information for the subtask process to solve each subtask. Since the subtask description step is a long-term task, in order to reduce the illusion when using a large model and obtain satisfactory component description information at the same time. The embodiment of the present invention adopts a combination of thinking chain and RAG. By querying and retrieving each step of the subtask description step, the most relevant related component information is obtained for each subtask. After obtaining all the relevant component information, the task description, subtask description steps, and related component information are input into the large model to obtain the final component description.

[0096] As an optional implementation, the step of retrieving each subtask description step to obtain a plurality of related component information can be specifically divided into the following steps S1021 to S1024:

[0097] S1021. Obtain a component library, where the component library contains multiple components.

[0098] S1022, extracting the semantic-aware embedding vector and the query embedding vector of each component in the component library;

[0099] S1023, obtaining the correlation between the subtask description step and the component according to the semantic perception embedding vector and the query embedding vector;

[0100] S1024: Determine the component with the highest correlation corresponding to each subtask description step, and obtain multiple pieces of relevant component information.

[0101] It should be noted that RAG can alleviate the illusion of large models. However, traditional RAG retrieves all relevant information at one time. For complex tasks such as generating corresponding component description information for subtask processes, it will be difficult to convert it into an effective search query, that is, it is difficult to predict the information required for subsequent reasoning and generation steps. At the same time, the task prompt itself is hardly enough to provide enough information to guide this process. This makes it difficult to find relevant documents, thereby limiting the scope of application of RAG. The thinking chain aims to improve the performance of large models in tasks that require complex steps, and to solve complex problems through a step-by-step thinking method. Therefore, an embodiment of the present invention adopts a method combining RAG and thinking chain to obtain the most correct component description.

[0102] Specifically, given a component library RAG aims to retrieve the most relevant components according to a query, where the query is described by subtasks in steps a=a1; a1; ...a T First, we extract the semantic-aware embedding vector of the component description in the component library. (i.e. each component C i The embedding representation of , a real vector of size K) and the embedding vector of the query Wherein, emb(·) represents an embedding function, which can be implemented using various text embedding models. In the embodiment of the present invention, Sentence-BERT is used.

[0103] Next, through the cosine similarity sin(Q,C i ) is used to measure the relevance of the subtask description steps and components, as shown in the following formula:

[0104]

[0105] Based on their relevance, the component with the highest relevance is selected, and then the corresponding component is retrieved for each subtask description step a to obtain the relevant component library C ‘ , related component library C ‘ Includes information about multiple related components.

[0106] Further, get the relevant component library C ‘ Then, the task requirement q and subtask description steps a = a1; a1; ... a T 、Related component library C ‘ As an input prompt, the big model generates the final component description information. The component description information is reviewed by the user. If the user confirms that the component description information is accurate, the corresponding component information will be included in the component library to enrich its content. If any problems are found, they will be fed back to the big model in a timely manner so that the answer can be regenerated and optimized.

[0107] S103, construct a prompt template, and obtain the workflow according to the prompt template, task requirements and component description information through the large model.

[0108] It should be noted that the large model needs to comprehensively consider the specific requirements in the task requirements and the parameter transfer between subtasks to build a complete workflow. This step is mainly completed based on the subtask description steps and component description information obtained above. Figure 2 The figure shows a schematic diagram of the component assembly process. The component assembly generation workflow is completed in two stages. The first stage is to provide an initial prompt for the large model, introduce the task context and available components, and generate the expected format of the workflow. The second stage is to use the components to generate subsequent workflows and deliver them to the user.

[0109] As an optional implementation, the step of constructing a prompt template can be specifically divided into the following steps S1031:

[0110] S1031, constructing a prompt template according to the component description information and the subtask description steps;

[0111] Among them, the prompt template includes role and target prompts, context information prompts, component description prompts, restriction prompts and demonstration prompts.

[0112] Specifically, in the first stage, the large model is taught knowledge through prompt templates so that it can understand and perform specific tasks. The prompt templates include the following five parts: role and target prompts, context information prompts, component description prompts, restriction prompts, and demonstration prompts.

[0113] (1) Role and target tips:

[0114] Outlining the large model roles and goals through role and goal cues provides clarity to the large model about the specific tasks it is expected to perform. By concisely defining roles and goals, intent and expectations can be more effectively conveyed with fewer symbols. This provides two benefits: First, it enables the large model to better conform to general requirements and produce output that conforms to the expected roles and goals. Second, concise text concentrates meaning within fewer symbols, allowing the large model to infer stronger relevant connections between text elements at higher attention weights while preventing dilution of the model's attention.

[0115] (2) Contextual information prompts:

[0116] The contextual information hint aims to provide the descriptive information about the specific system (i.e., the subtask description step) that the large model needs to perform effective reasoning. The information provided in this context section serves two purposes: first, it enables the large model to understand the operation of the system and incorporate additional information about the specific system. Second, since the large model has been trained on a wide range of data, it possesses a large amount of implicit general knowledge stored in its model weights. The descriptive information provided in the hint guides the large model to "focus" on the relevant knowledge embedded in the model when generating text. In this sense, this process achieves the combination of the general knowledge of the large model with the specific knowledge specified in the hint to perform reasoning for a specific task.

[0117] (3) Component description tips:

[0118] The component description prompt provides all component descriptions needed to complete the task requirement q. The large model needs to complete the assembly generation workflow based on these component descriptions.

[0119] (4) Restriction Tips:

[0120] Restriction prompts limit the output of large models by specifying the required output format and generated content. In the embodiment of the present invention, the JSON structure is used to organize the input / output data of all operations in the workflow. The input data is presented in the form of key-value pairs, and each data has a specific key for easy parsing and operation. When transferring data between different software, the JSON structure is convenient for indexing specific data fields. Through the way of thinking chain, logical structured reasoning is promoted to ensure that the format of the generated workflow is correct.

[0121] (5) Demonstration Tips:

[0122] Demonstration prompts provide proven, concrete examples that demonstrate the desired input-output pattern. This can be beneficial in several ways: First, it constrains the structure of the text to be generated. Second, the performance of the model can be improved even with a limited number of examples. Finally, examples can help further clarify contextual information and disambiguate the abstract information provided so far.

[0123] It should be noted that the embodiment of the present invention provides the big model with the task background, detailed description and demonstration of available components through the prompt template, thereby helping the big model to understand the task objectives, which can effectively improve the understanding and execution capabilities of the big model, thereby improving the efficiency and accuracy of subsequent workflow generation.

[0124] As an optional implementation, the step of obtaining the workflow according to the prompt template, task requirements and component description information through the large model can be specifically divided into the following steps S1032 and S1033:

[0125] S1032, inputting the prompt template and the task requirement into the large model to obtain a workflow description file;

[0126] S1033. Obtain user feedback results according to the workflow description file, and update the workflow description file according to the user feedback results through the large model to obtain the workflow.

[0127] Specifically, in the second stage, the big model uses the knowledge obtained from the component description in the first stage and a deep understanding of user needs to automatically generate a structured and executable workflow description file. This workflow description file contains the steps and logic of the workflow, which can be directly imported into the low-code platform for execution. The big model shows the user a high-level description of the generated workflow (i.e., the workflow description file), which explains the logic and function of the workflow in concise and clear language without the need for the user to deeply understand the underlying code. The user can provide user feedback results based on the workflow description file, such as pointing out errors, making suggestions for improvement, or clarifying issues. Then the big model adjusts the workflow description file according to the user feedback results, and makes necessary improvements to ensure that it accurately meets user needs and obtains the workflow that is ultimately used for execution on the low-code platform. The embodiment of the present invention enhances the two-way interaction between the big model and the user through the above-mentioned user feedback mechanism, which can improve the accuracy and reliability of workflow generation and enhance user trust and satisfaction with the system.

[0128] The above is a description of the cross-system business integration workflow generation method based on a large model according to an embodiment of the present invention. It can be appreciated that, compared with the workflow generation method in the prior art, the embodiment of the present invention has the following advantages:

[0129] 1. Automatically decompose the task requirements described by the user into a series of operable subtask description steps through the big model, and automatically retrieve the most relevant component description information based on the subtask description steps, and finally automatically generate a structured and executable workflow, which can realize the efficient generation of cross-system business process component fusion workflow;

[0130] Second, we introduced reinforcement learning and heuristic search techniques to guide the large model to select the optimal subtask execution path, and introduced RAG technology to reduce the probability of errors and hallucinations in the model, thereby improving the accuracy and reliability of the workflow;

[0131] 3. Through the user feedback mechanism, users are allowed to adjust and optimize the generated workflow so that the generated workflow can more accurately meet user needs and can be flexibly adjusted according to user needs to adapt to different business scenarios.

[0132] Reference Figure 3The embodiment of the present invention further provides a cross-system business fusion workflow generation system based on a large model, including:

[0133] The task decomposition module is used to obtain task requirements and decompose them through a large language model to obtain multiple subtask description steps;

[0134] The component generation module is used to retrieve each subtask description step to obtain multiple related component information, and then input the task requirements, subtask description steps and related component information into a preset large model to obtain component description information;

[0135] The workflow generation module is used to build a prompt template and obtain the workflow according to the prompt template, task requirements and component description information through the large model.

[0136] The contents of the above-mentioned cross-system business fusion workflow generation method based on big model embodiment are all applicable to the present cross-system business fusion workflow generation system embodiment based on big model. The functions specifically implemented by the present cross-system business fusion workflow generation system embodiment based on big model are the same as those of the above-mentioned cross-system business fusion workflow generation method based on big model embodiment, and the beneficial effects achieved are also the same as those achieved by the above-mentioned cross-system business fusion workflow generation method based on big model embodiment.

[0137] The embodiment of the present invention also provides an electronic device, the electronic device comprising: a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for realizing connection and communication between the processor and the memory, and when the program is executed by the processor, the above-mentioned cross-system business fusion workflow generation method based on a large model is realized. The electronic device can be any intelligent terminal including a tablet computer, a vehicle-mounted computer, etc.

[0138] like Figure 4 FIG. 1 is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention, referring to FIG. Figure 4 , an embodiment of the present invention provides an electronic device, including:

[0139] The processor 1001 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present invention.

[0140] The memory 1002 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1002 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 1002, and the processor 1001 calls and executes the cross-system business fusion workflow generation method based on a large model in the embodiment of the present invention;

[0141] Input / output interface 1003, used to implement information input and output;

[0142] The communication interface 1004 is used to realize the communication interaction between the device and other devices. The communication can be realized through a wired manner (such as USB, network cable, etc.) or a wireless manner (such as mobile network, WIFI, Bluetooth, etc.);

[0143] A bus 1005 , which transmits information between various components of the device (e.g., the processor 1001 , the memory 1002 , the input / output interface 1003 , and the communication interface 1004 );

[0144] The processor 1001 , the memory 1002 , the input / output interface 1003 and the communication interface 1004 are connected to each other in communication within the device via the bus 1005 .

[0145] An embodiment of the present invention also provides a storage medium, which is a computer-readable storage medium used for computer-readable storage. The storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the above-mentioned large model-based cross-system business fusion workflow generation method.

[0146] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0147] The embodiment of the present invention also discloses a computer program product or a computer program, wherein the computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes Figure 1 The method shown.

[0148] In some selectable embodiments, the function / operation mentioned in the block diagram may not occur in the order mentioned in the operation diagram. For example, depending on the function / operation involved, the two boxes shown in succession can actually be executed substantially simultaneously or the above-mentioned boxes can sometimes be executed in reverse order. In addition, the embodiment presented and described in the flow chart of the present invention is provided by way of example, for the purpose of providing a more comprehensive understanding of technology. The disclosed method is not limited to the operation and logic flow presented herein. Selectable embodiments are expected, wherein the order of various operations is changed and the sub-operation of a part for which is described as a larger operation is performed independently.

[0149] In addition, although the present invention is described in the context of functional modules, it should be understood that, unless otherwise specified to the contrary, one or more of the above-mentioned functions and / or features can be integrated into a single physical device and / or software module, or one or more functions and / or features can be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the present invention. More specifically, in view of the properties, functions and internal relationships of the various functional modules in the device disclosed herein, the actual implementation of the module will be understood within the conventional skills of the engineer. Therefore, those skilled in the art can implement the present invention set forth in the claims without excessive experimentation using ordinary techniques. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present invention, which is determined by the full scope of the appended claims and their equivalents.

[0150] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the above methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0151] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0152] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0153] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

[0154] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. A method for generating a cross-system business integration workflow based on a large model, characterized in that: The following steps are involved: Obtaining task requirements, decomposing the task requirements through a preset large model, and obtaining multiple subtask description steps; Retrieving each of the subtask description steps to obtain a plurality of related component information, and then inputting the task requirements, the subtask description steps and the related component information into the large model to obtain component description information; A prompt template is constructed, and a workflow is obtained according to the prompt template, the task requirements and the component description information through the large model.

2. According to claim 1, a method for generating a cross-system business integration workflow based on a large model is characterized in that: Decomposing the task requirements by using a preset large model to obtain multiple subtask description steps specifically includes: Determine the optimal strategy of the large model and input the task requirements into the large model; Obtaining multiple states according to the task requirements, and obtaining cumulative benefits and heuristic values ​​according to all the states; Calculating the expansion value corresponding to each of the states according to the accumulated income and the heuristic value; Sort the states from high to low according to the extension value, and determine a preset number of the states as candidate states according to the sorting result; Each of the candidate states is expanded according to the optimal strategy to obtain a plurality of subtask description steps.

3. According to claim 2, a method for generating a cross-system business integration workflow based on a large model is characterized in that: The multiple states are obtained according to the task requirements, specifically: The decomposition process of the task requirements is conceptualized as a Markov decision framework, which includes a plurality of states, a plurality of actions, a discount factor, and a reward function for a state-action pair.

4. According to claim 3, a method for generating a cross-system business integration workflow based on a large model is characterized in that: The cumulative benefits and heuristic values ​​are obtained according to all the states, specifically including: Calculate the rewards of all the states, perform aggregation operations on the rewards, and obtain the cumulative income; The optimal cumulative expected reward value of the state-action pair is calculated according to the reward function and the discount factor, and the optimal cumulative expected reward value is used as the heuristic value.

5. According to the method for generating a cross-system business integration workflow based on a large model according to claim 1, it is characterized in that: The step of retrieving each of the subtask descriptions to obtain a plurality of related component information specifically includes: Obtain a component library, wherein the component library includes multiple components; Extracting a semantically-aware embedding vector and a query embedding vector of each of the components in the component library; Obtaining the relevance of the subtask description step and the component according to the semantically-aware embedding vector and the query embedding vector; The component with the highest correlation corresponding to each of the subtask description steps is determined to obtain a plurality of related component information.

6. The method for generating a cross-system business integration workflow based on a large model according to claim 1 is characterized in that: The construction prompt template is specifically: Constructing the prompt template according to the component description information and the subtask description steps; The prompt template includes role and target prompts, context information prompts, component description prompts, restriction prompts and demonstration prompts.

7. The method for generating a cross-system business integration workflow based on a large model according to claim 1 is characterized in that: The obtaining of the workflow according to the prompt template, the task requirement and the component description information through the large model specifically includes: Input the prompt template and the task requirement into the large model to obtain a workflow description file; A user feedback result is obtained according to the workflow description file, and the workflow description file is updated according to the user feedback result through the large model to obtain the workflow.

8. A cross-system business integration workflow generation system based on a large model, characterized in that: include: A task decomposition module is used to obtain task requirements and decompose the task requirements through a large language model to obtain multiple subtask description steps; A component generation module, used to retrieve each of the subtask description steps to obtain a plurality of related component information, and then input the task requirements, the subtask description steps and the related component information into a preset macro model to obtain component description information; The workflow generation module is used to construct a prompt template, and obtain the workflow according to the prompt template, the task requirements and the component description information through the large model.

9. An electronic device, characterized in that: The electronic device includes a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for realizing connection and communication between the processor and the memory. When the program is executed by the processor, the steps of the large model-based cross-system business fusion workflow generation method as described in any one of claims 1 to 7 are realized.

10. A storage medium, the storage medium being a computer-readable storage medium, used for computer-readable storage, characterized in that: The storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of the large model-based cross-system business fusion workflow generation method as described in any one of claims 1 to 7.

Images