Method, device and equipment for training workflow to create model, medium and product

Abstract

The embodiment of the invention provides a method, device and equipment for training a workflow to create a model, a storage medium and a program product. The method comprises the steps of obtaining a creation request for creating a workflow for a data object; acquiring data of a workflow corresponding to the creation request, wherein the workflow is created based on the creation request and the information of the data object; based on the data of the workflow, determining first sample data related to a workflow framework of the workflow and second sample data related to configuration information of a plurality of nodes in the workflow framework; and training a workflow creation model based on the first sample data and the second sample data. A workflow creation task is divided into two types of sub-tasks, namely, a framework generation sub-task and a node configuration generation sub-task, so that the learning target of the model in each training stage can be clarified. In this way, the adaptation capability and creation quality of the workflow creation model in a real business environment can be improved.

Description

Technical Field

[0001] The exemplary embodiments disclosed herein generally relate to the field of computers, and more particularly to methods, apparatus, electronic devices, computer-readable storage media, and computer program products for training workflow creation models. Background Technology

[0002] With the rapid development of machine learning and data analytics technologies, an increasing number of office and management systems are introducing intelligent assistants to simplify the creation and execution of tasks involving data objects. For example, workflow creation solutions based on machine learning models are gradually becoming a means to improve efficiency and reduce labor costs. However, model-based workflows may not fully meet user needs, especially when the process logic is complex and the data scale is large. Therefore, how to train the creation model to improve the quality of the created workflows has become a noteworthy issue. Summary of the Invention

[0003] In a first aspect of this disclosure, a method for training a workflow creation model is provided. The method includes: acquiring a creation request for creating a workflow for a data object; acquiring workflow data corresponding to the creation request, the workflow being created based on the creation request and information about the data object; determining, based on the workflow data, first sample data related to the workflow framework and second sample data related to configuration information of multiple nodes in the workflow framework; and training a workflow creation model based on the first and second sample data.

[0004] In a second aspect of this disclosure, an apparatus for training a workflow creation model is provided. The apparatus includes: a creation request acquisition module configured to acquire a creation request for creating a workflow based on a data object; a workflow data acquisition module configured to acquire workflow data corresponding to the creation request, wherein the workflow is created based on the creation request and information about the data object; a sample data determination module configured to determine, based on the workflow data, first sample data related to the workflow framework and second sample data related to configuration information of multiple nodes in the workflow framework; and a training module configured to train the workflow creation model based on the first and second sample data.

[0005] In a third aspect of this disclosure, an electronic device is provided. The device includes at least one processor; and at least one memory coupled to the at least one processor and storing instructions for execution by the at least one processor. When executed by the at least one processing unit, the instructions cause the electronic device to perform the method of the first aspect.

[0006] In a fourth aspect of this disclosure, a computer-readable storage medium is provided. The medium stores computer instructions that, when executed by a processor, implement the method of the first aspect.

[0007] In a fifth aspect of this disclosure, a computer program product is provided. The product includes a computer program, wherein when executed by a processor, the computer program implements the method according to a first aspect of this disclosure.

[0008] It should be understood that the content described in this content section is not intended to limit the key or essential features of the embodiments of this disclosure, nor is it intended to restrict the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0009] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. In the drawings, the same or similar reference numerals denote the same or similar elements, wherein:

[0010] Figure 1 A schematic diagram of an example environment in which embodiments of the present disclosure can be implemented is shown;

[0011] Figure 2 A schematic diagram illustrating an example process for creating a workflow according to some embodiments of this disclosure is shown;

[0012] Figure 3 A schematic diagram of an example interface for creating a workflow according to some embodiments of the present disclosure is shown;

[0013] Figure 4 A schematic diagram illustrates an example process for creating a model using a training workflow according to some embodiments of the present disclosure;

[0014] Figure 5 A flowchart is shown illustrating a method for training a workflow creation model according to some embodiments of the present disclosure;

[0015] Figure 6 A schematic structural block diagram of an apparatus for training a workflow creation model according to some embodiments of the present disclosure is shown; and

[0016] Figure 7 A block diagram of an electronic device in which one or more embodiments of the present disclosure may be implemented is shown. Detailed Implementation

[0017] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0018] In the description of embodiments of this disclosure, the term "comprising" and similar terms should be understood as open-ended inclusion, i.e., "including but not limited to". The term "based on" should be understood as "at least partially based on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The term "some embodiments" should be understood as "at least some embodiments". Other explicit and implicit definitions may also be included below.

[0019] In this document, unless explicitly stated otherwise, performing a step in response to A does not mean that the step is performed immediately after A, but may include one or more intermediate steps.

[0020] It is understood that the data involved in this technical solution (including but not limited to the data itself, the acquisition, use, storage or deletion of the data) shall comply with the requirements of relevant laws, regulations and related provisions.

[0021] It is understood that before using the technical solutions disclosed in the various embodiments of this disclosure, relevant users should be informed of the type, scope of use, and usage scenarios of the information involved in this disclosure through appropriate means in accordance with relevant laws and regulations, and authorization should be obtained from the relevant users. Among them, relevant users may include any type of rights holder, such as individuals, enterprises, and groups.

[0022] For example, in response to receiving an active request from a user, a prompt message is sent to the relevant user to clearly inform the user that the requested operation will require obtaining and using the user's information, thereby enabling the relevant user to choose whether to provide information to the software or hardware such as the electronic device, application, server, or storage medium that performs the operation of the technical solution disclosed herein based on the prompt message.

[0023] As an optional but non-restrictive implementation, in response to a user's active request, a prompt message can be sent to the user, such as a pop-up window, where the prompt message can be presented in text format. Furthermore, the pop-up window can also include a selection control allowing the user to choose "agree" or "disagree" to provide information to the electronic device.

[0024] It is understood that the above notification and user authorization process are merely illustrative and do not constitute a limitation on the implementation of this disclosure. Other methods that comply with relevant laws and regulations may also be applied to the implementation of this disclosure.

[0025] As used herein, the term "model" refers to a model that learns the relationship between inputs and outputs from training data, enabling it to generate corresponding outputs for a given input after training. Model generation can be based on machine learning techniques. Deep learning is a machine learning algorithm that processes inputs and provides corresponding outputs using multiple layers of processing units. A neural network model is an example of a deep learning-based model. In this document, "model" may also be referred to as a "machine learning model," "learning model," "machine learning network," or "learning network," and these terms are used interchangeably. In some embodiments described below, the machine learning model may be a large language model or have an architecture based on a large language model. A machine learning model can also be a multimodal model capable of handling multimodal inputs (e.g., text input and visual input).

[0026] Figure 1 A schematic diagram of an example environment 100 in which embodiments of the present disclosure can be implemented is shown. Figure 1 In environment 100, it is desired to train and use a machine learning model (i.e., model 130) configured for various application environments. For example, model 130 is configured to assist in building workflows to support various data object-based business automation scenarios.

[0027] like Figure 1 As shown, environment 100 includes model training system 150 and model application system 160. Figure 1 The upper part illustrates the model training phase, and the lower part illustrates the model application phase. Before training, the parameter values ​​of model 130 can have initial values ​​or pre-trained parameter values ​​obtained through a pre-training process. Model 130 can be trained via forward and backward propagation, during which the parameter values ​​of model 130 can be updated and adjusted. After training is complete, model 130' is obtained. At this point, the parameter values ​​of model 130' have been updated, and based on the updated parameter values, model 130 can be used to implement workflow creation tasks in the model application phase.

[0028] During the model training phase, the model 130 can be trained using a training sample set 110 comprising multiple training samples 112 and a model training system 150. The training samples 112, including model input 120 and model output 122, can be used to train the model 130. For example, model input 120 may include natural language text representing the target business intent, data table information related to the business system, etc., and model output 122 may include the corresponding workflow. Specifically, the training process can be executed iteratively using a large number of training samples. After training is complete, the model 130 may include knowledge about the task to be processed.

[0029] During the model application phase, model 130' (which now has trained parameter values) can be used to perform the corresponding task. Specifically, model application system 160 can receive model input 142 and output corresponding model output 144. For example, model application system 160 can be embedded in a digital assistant. Users can submit requests to create workflows by accessing the digital assistant. Based on request 142, model application system 160 can assist users in quickly creating data object-based workflows 144.

[0030] exist Figure 1 In this context, the model application system 160 may include any computing system with computing capabilities, such as various computing devices / systems, terminal devices, servers, etc. Terminal devices may involve any type of mobile terminal, fixed terminal, or portable terminal, including mobile phones, desktop computers, laptop computers, notebook computers, netbook computers, tablet computers, media computers, multimedia tablets, or any combination thereof, including accessories and peripherals of these devices or any combination thereof. Servers include, but are not limited to, mainframes, edge computing nodes, computing devices in cloud environments, etc.

[0031] It should be understood that the structure and function of the various elements in environment 100 are described for illustrative purposes only and do not imply any limitation on the scope of this disclosure.

[0032] As briefly mentioned earlier, with the development of data analysis and machine learning technologies, more and more workflow processing systems are exploring ways to assist users in creating automated task processes in a smarter and more accessible manner. In some business scenarios, users need to set specific processing logic based on data records, such as triggering notifications, conditional branching, and data aggregation. To address this, some platforms have begun to provide low-code or graphical workflow editing tools, enabling users to configure automated processes by dragging and dropping nodes and filling in parameters.

[0033] In related technologies, some platforms have begun exploring the application of machine learning models to the automated workflow construction process, attempting to guide the model to automatically create complete workflows through prompts. However, workflow protocols themselves are complex in structure, especially in scenarios involving data objects (such as multidimensional data tables, dashboards, etc.), often containing multiple node types, numerous field configurations, and nested parameter logic. In such cases, relying solely on prompts to create workflows is unlikely to achieve reliable and stable results, failing to meet the requirements of process completeness and execution accuracy in real business scenarios.

[0034] In view of this, according to embodiments of the present disclosure, an improved scheme for training a workflow creation model is provided. According to this scheme, firstly, a creation request for creating a workflow based on a data object is obtained. Next, workflow data corresponding to the creation request can be obtained, the workflow being created based on the creation request and information about the data object. Further, based on the workflow data, first sample data related to the workflow framework and second sample data related to the configuration information of multiple nodes in the workflow framework can be determined. Still further, a workflow creation model can be trained based on the first and second sample data.

[0035] Therefore, by acquiring the workflow creation request for the data object and the data of the workflow created based on that request, high-quality training samples that meet the expected goals can be obtained. Based on this, the workflow creation task is further divided into two sub-tasks: generating the framework and configuring the generating nodes. Sample data and training processes are designed for these two sub-tasks in the data construction and training phases, respectively. This approach clarifies the learning objectives of the model at each training phase. This method helps improve the adaptability and creation quality of the workflow creation model in real-world business environments.

[0036] The following description continues with reference to the accompanying drawings, outlining some exemplary embodiments of this disclosure. Due to the high complexity of various workflow protocols, especially those involving multiple node types, numerous field bindings, and nested parameter configurations, workflow creation models often struggle to directly generate a complete and accurate workflow within a single stage. Therefore, in some embodiments, the workflow creation process can be divided into two stages, with separate training models for each stage and training tasks designed for each stage. The first stage can focus on generating the overall structure of the workflow, i.e., the workflow framework. The second stage, based on the workflow framework, can generate the configuration information for each node in the workflow framework node by node, thereby completing the specific execution logic and parameter settings for each node. This division strategy not only reduces the difficulty of model training but also allows each stage to focus on specific sub-tasks, thereby improving the overall generation quality and stability.

[0037] Figure 2 A schematic diagram of an example process 200 for workflow creation according to some embodiments of this disclosure is shown. Some or all of the process 200 may be implemented by the model application system 160 or by other devices, such as other remote devices (terminal devices or service devices) with computing capabilities. It should be understood below that actions described relative to the model application system 160 may be performed at least partially by entities other than the model application system 160, such as by a digital assistant, or by the model application system 160 alone, by the server 130 alone, or by both in conjunction.

[0038] like Figure 2 As shown, the model application system 160 can receive workflow creation requests originating from data objects. A data object refers to a data entity in various forms used to carry business information. Data objects can include structured data objects (e.g., data tables, database tables, etc.) and unstructured data objects. Data objects can be presented in the interface in the form of tables, forms, or charts. For example, a data object can include one or more multidimensional data tables with related structural information. Each data object can contain a field structure and data records. The field structure includes the number of fields, field names, field types, etc., and the data records correspond to the actual data rows in the table.

[0039] In some embodiments, a data object can be associated with multiple functional components, such as forms for data input, dashboards for data analysis, and so on. The dashboard may include statistical analysis results of the data table content and display them in various chart formats, such as bar charts, line charts, and pie charts. When users manipulate these data objects—for example, when viewing tables, editing forms, or analyzing charts—they can initiate workflow creation or modification requests based on the data content.

[0040] A workflow is an execution logic chain composed of multiple functional nodes. These nodes are organized according to a specific topology and are used to process business logic associated with data objects. The design, configuration, and triggering of a workflow can all depend on information from one or more data objects. The core function of a workflow is to achieve data-driven automated operations.

[0041] In some embodiments, data objects can provide contextual information and data sources for workflow creation and configuration. The workflow's structure and node configuration can be based on the field structure, field value types, and historical records in the data object. For example, when a user wants to set up a process to "notify the responsible person when the project status is delayed," the model application system 160 can identify the "project status" field contained in the data object and generate corresponding decision and notification node configurations accordingly.

[0042] Alternatively or additionally, workflow execution can target data records within a data object. For example, a decision node can iterate through each record in a data table, triggering subsequent operations on rows that meet certain conditions. The execution results can also be written back to the original data object or its associated state updated. In some embodiments, a data object can be mapped to multiple workflow instances. Different workflows can operate on different fields, record sets, or business views of the data object, thereby supporting multi-dimensional automation capabilities.

[0043] refer to Figure 2 In box 210, the model application system 160 can perform requirement identification based on the workflow editing request to determine whether the user's input is a workflow creation request that can be processed. For example, a workflow creation request may indicate that the user wants to create a new workflow or make structural adjustments or configuration modifications to an existing workflow. If the creation request is relevant to the workflow and clearly expressed, the model application system 160 can execute box 220 to generate the workflow framework. Further, the model application system 160 can execute box 225 to generate configuration information for the nodes in the workflow framework. During this process, the model application system 160 can execute box 230 to determine whether the current node is the last node in the workflow framework. If the current node is the last node in the workflow framework, the model application system 160 can create workflow 231 based on the generated workflow framework and configuration information. In box 240, the model application system 160 can render workflow 231 and present it to the user.

[0044] As an example, Figure 3 A schematic diagram of an example interface 300 for creating a workflow according to some embodiments of the present disclosure is shown. Figure 3 As shown, the model application system 160 can receive workflow creation requests via input box 301. The received creation request can be presented in the user interface as a dialog box 302, allowing the user to view historical input. Based on the received creation request, the model application system 160 can generate a workflow framework 303. Furthermore, the model application system 160 can determine the configuration information of each node in the workflow framework 303. For example, as presented by the information configuration component 304, the model application system 160 can configure fields such as sender and receiver, as well as information such as the message title and content, for the "Send Message" node. It should be understood that... Figure 3 The workflow framework 303 in the diagram is merely illustrative. The workflow framework 303 may include more, fewer, or different nodes, and may include different branches or different processes.

[0045] Therefore, the workflow data mainly includes workflow framework data describing the workflow structure, and configuration data specifying the behavior and operation of each node. Thus, to implement the aforementioned workflow creation process, the model training system 150 can construct corresponding training samples to train the workflow creation model 130, enabling it to generate workflow frameworks and configuration information.

[0046] Figure 4 A schematic diagram of an example process 400 for creating a model using a training workflow according to some embodiments of this disclosure is shown. Some or all of the process 400 may be implemented by the model training system 150 or by other devices, such as other remote devices (terminal devices or service devices) with computing capabilities. It should be understood below that the actions described relative to the model training system 150 may be performed at least partially by entities other than the model training system 150, for example, by the model training system 150 alone, by other terminal devices or servers alone, or by both in conjunction.

[0047] In some embodiments, the model training system 150 can acquire creation requests for creating workflows based on data objects. The model training system 150 can acquire multiple creation requests for workflows constructed based on business scenarios. The creation request can be a description of the process goals the user wishes to achieve, expressed in natural language.

[0048] In some embodiments, a creation request may correspond to multiple different business scenarios. The model training system 150 can acquire multiple creation requests to construct multiple training samples covering different scenarios. The model training system 150 can construct a set of training sample data based on each creation request, thereby providing the model with diverse learning materials.

[0049] In some embodiments, the model training system 150 can acquire requirement information associated with the creation workflow. Further, the model training system 150 can generate a creation request based on the requirement information. (See reference...) Figure 4 In box 401, model training system 150 can collect user requests. For example, model training system 150 can acquire natural language requests entered by users when creating workflows, which may come from online logs, business research documents, or simulated human input.

[0050] In box 402, the model training system 150 can determine demand information based on the collected user needs. For example, the model training system 150 can perform classification and statistical analysis on existing user requests to uncover data features and common patterns. For instance, the model training system 150 can determine information such as differences in the expression of different user requests, frequently mentioned node types, and the average number of nodes involved in each creation request. In some embodiments, the model training system 150 can combine business expectations and demand information to identify representative request types (such as "approval flow requests," "data reminder requests," etc.) and thereby summarize the direction of creation requests required to construct training samples. In this way, a systematic planning of the training sample construction strategy can be achieved, thereby ensuring that the generated training data covers typical scenarios and meets the requirements of diversity and accuracy for subsequent model training.

[0051] Alternatively and / or additionally, in box 403, the model training system 150 can determine quality requirements based on the collected user needs. For example, for a workflow creation request of the "send message" type, the quality requirements might indicate that certain field conditions in the request must be specific and that the sending action must specify the sender and receiver. The process of collecting user needs and determining the need information and quality requirements can be accomplished using an intent recognition model, a classification model, or other algorithmic models. Based on this, the model training system 150 can determine the key directions and variation space that subsequent training samples need to cover, thereby providing a basis for the construction of subsequent training data.

[0052] Continue to refer to Figure 4 In box 411, the model training system 150 can obtain a creation request for constructing training data. The creation request can be manually constructed based on the aforementioned defined requirements. For example, the model training system 150 can obtain a natural language request text (i.e., ground truth data) written according to preset rules such as construction direction, quality requirements, and node types. Alternatively and / or additionally, the creation request can be automatically generated by a pre-trained model and then manually filtered and labeled.

[0053] In box 412, the model training system 150 can determine whether the creation request meets the quality requirements associated with the creation workflow. If the creation request does not meet the quality requirements (e.g., incomplete structure, unclear semantics, or missing field), the model training system 150 can adjust the creation request. For example, the model training system 150 can supplement missing node information in the request, adjust the request expression, or obtain a new creation request.

[0054] In some embodiments, the model training system 150 can acquire workflow data corresponding to the creation request. The workflow is created based on the creation request and information about the data object. (See reference) Figure 4 In box 413, the model training system 150 can create a workflow corresponding to each creation request. Each workflow is created based on the information of the corresponding creation request and data object. For example, a workflow may include one or more created through a graphical interface based on the information of the creation request and data object, such as a first workflow, a second workflow, and so on.

[0055] As an example, annotators can select node types, configure node parameters, and build a complete workflow in a graphical interface based on information such as the target, operation object, judgment conditions, and execution actions in the creation request. The workflow created in this way not only covers the functional nodes and execution logic required for the creation request but also includes the configuration information and execution order of each node. The extracted structural and parameter information can serve as a core component of the training samples.

[0056] In box 414, the model training system 150 can determine whether the first workflow matches the creation request. For example, the model training system 150 can verify whether the structure of the workflow framework of the first workflow covers all the intent expressed in the request, whether the function of the nodes is consistent with the goal expressed by the user, and whether the execution path can cover the business process implicit in the request, thereby determining whether the first workflow matches the creation request.

[0057] In some embodiments, if the first workflow matches the creation request, process 400 can proceed to block 415. In block 415, the model training system 150 can extract data from the first workflow according to a predetermined data format. The model training system 150 can extract structured data from the first workflow according to a predetermined data format or data structure. For example, the model training system 150 can extract data such as the type and connection relationships of each node in the workflow, and the configuration parameters (field bindings, action settings) of each node.

[0058] In some embodiments, if the first workflow does not match the creation request, the model training system 150 can extract data from the second workflow according to a predetermined data format. The second workflow is recreated through a graphical interface based on the creation request and data object information. For example, the second workflow could be a version reconstructed manually based on the same creation request and data object information. In this way, the sample data used for training can be highly consistent and of high quality, thereby improving the generation accuracy and generalization ability of the workflow creation model in real business scenarios.

[0059] In some embodiments, after multiple training samples have been constructed, the model training system 150 can determine whether the number of current sample data meets the preset training requirements. If the number or distribution of sample data does not meet the training requirements, the model training system 150 can continue to obtain more creation requests and execute the same workflow creation and data extraction process to supplement the coverage and representativeness of the training dataset.

[0060] Continue to refer to Figure 4 In box 421, the model training system 150 can determine sample data. In some embodiments, the model training system 150 can determine, based on workflow data, first sample data related to the workflow framework and second sample data related to the configuration information of multiple nodes in the workflow framework.

[0061] In some embodiments, the model training system 150 may determine first sample data based on workflow data. The first sample data may include the topology of multiple nodes, the types of nodes among the multiple nodes, and at least one operation of the nodes. For example, Table 1 shows an example of the first sample data.

[0062] Table 1 Example of the first sample data

[0063]

[0064]

[0065]

[0066]

[0067] As shown in Table 1, the first sample data may include input data related to the generated workflow framework. For example, the first sample data may include a creation request, i.e., a user's description of the workflow requirements, used to drive the model to understand the core logical chain that the entire workflow should include. The first sample data may include information about data objects, such as information from several multidimensional data tables related to the business system. This table information serves as contextual input, helping the model understand fields such as "Task ID," "Original Completion Time," "Responsible Person," and "Work Content" at the field level, and correctly locate the fields and logical relationships when generating the process structure.

[0068] Referring back to Table 1, the first sample data may include the model's output data. In the training task corresponding to the first sample data, the model's output target is a workflow framework. The workflow framework may include the topology of nodes, as well as the basic order and type of process nodes, but the specific configuration information of each node (such as "data:{}") is empty. The first sample data can be used to train a workflow creation model to generate a workflow framework.

[0069] In some embodiments, the model training system 150 may determine second sample data based on workflow data and at least one operation of a node. The second sample data includes parameters of at least one operation of the node and the numerical values ​​of those parameters. For example, Table 2 shows an example of the second sample data.

[0070] Table 2 Examples of the second sample data

[0071]

[0072]

[0073]

[0074]

[0075]

[0076] As shown in Table 2, the second sample data may include input data related to the configuration information of a specific node in the workflow. This type of sample data can be used to train the model to understand the context of each node in the entire process, and, combined with user requirements, data object structure, and table view information, output configuration information that meets business semantics. The second sample data may include the overall workflow creation request, configuration requirements for the current node, data object information, current time, and the ID of the node to be configured. Alternatively or additionally, the second sample data may also include the current workflow draft, i.e., the generated framework and configuration information of the preceding nodes, thereby providing the model with a process structure context.

[0077] Referring again to Table 2, the second sample data may include parameters for at least one operation of the node, specifying which operation parameters the node needs to fill in, their names, data types, business meanings, and filling rules. The second sample data may also include the model's output data. In the training task corresponding to the second sample data, the model's output target is the node's configuration information (e.g., the numerical values ​​of parameters for at least one operation of the node).

[0078] In some embodiments, the second sample data may further include view information associated with the data object. The view information indicates at least one of the following: the filtering conditions or sorting methods for the information of the data object. For example, the view information may originate from different visualization configurations for one or more table boxes in a multidimensional table system. For instance, multiple users can define multiple views around the same "project task table," each view based on the same data source but with different filtering, sorting, and field display rules to meet their respective business concerns. Introducing view information associated with the data object into the second sample data helps the model more accurately configure the parameters of each process node in accordance with users' actual business usage habits, especially in scenarios involving complex configurations such as field referencing, filtering logic, and conditional judgments.

[0079] It should be understood that Tables 1 and 2 are merely illustrative, and different sample data can be determined for different workflows. Suppose the desired workflow is to look up the production data table and send a notification to the manager when output exceeds a predetermined threshold. In this case, the first sample data could specify multiple nodes for performing the following steps: periodically querying the production data table, determining the manager's communication address, and sending a message to the communication address when output exceeds the predetermined threshold, etc. The second sample data could include specific configuration information for each node, such as the access address of the production data table, the name of the output field, etc.

[0080] In some embodiments, the model training system 150 can create a model based on a training workflow using first sample data and second sample data. (Return to Reference) Figure 4 In box 422, the model training system 150 can construct training data based on two types of sample data: the first sample data and the second sample data. For example, the model training system 150 can process and combine these two types of sample data to meet the data format, field specifications, and task structure supported by the workflow creation model. Furthermore, in box 423, the model training system 150 can perform a phased model training process based on this training data to improve the workflow creation model's capabilities in both framework generation and node configuration dimensions.

[0081] In some embodiments, the model training system 150 can train a first model structure based on first sample data. The first model structure is used to generate predictions of the workflow framework. The first model structure can be understood as a model functional block that encapsulates the atomic capabilities of the workflow framework generation. By invoking these atomic capabilities, the workflow creation model can automatically complete the type inference, sequence arrangement, and output of the overall process structure of each node in the workflow based on the creation request and data object information.

[0082] In some embodiments, the model training system 150 can generate a prediction of the workflow framework by invoking a first model structure. For example, the model training system 150 can generate a prediction of the workflow framework based on a creation request, information about data objects, and a framework-generated prompt template. The creation request may include a workflow creation request expressed in natural language. The information about data objects may include multiple business-related data tables and their field information. The framework-generated prompt template is used to guide the model to better understand the input and generate the output of the workflow framework.

[0083] In some embodiments, the model training system 150 can train a first model structure based on the predictions from the workflow framework and the first sample data. The model training system 150 can also perform supervised training on the first model structure based on the differences between the prediction results and the corresponding first sample data. For example, the model training system 150 can iteratively improve the generation accuracy and structural rationality of the first model structure by optimizing the objective function.

[0084] In some embodiments, the model training system 150 can generate configuration information for multiple nodes in the workflow framework prediction by invoking a second model structure. The second model structure can be understood as a model function module that encapsulates the atomic capability of generating node parameters. By invoking this atomic capability, the model can complete the operation parameter values ​​for each node based on inputs such as node information, node descriptions, and data object backgrounds in the workflow structure, thereby completing the configuration process for each node in the entire workflow.

[0085] In some embodiments, the model training system 150 can generate prompt word templates for nodes among multiple nodes, at least based on the node's description information and configuration, and generate numerical values ​​of parameters for at least one operation of the node as at least part of the node's configuration information. For example, the model training system 150 can call a second model structure to configure multiple nodes in the workflow framework generated by the first model structure one by one to generate configuration information corresponding to each node.

[0086] In some embodiments, the model training system 150 can train a second model structure based on the configuration information of multiple nodes and the second sample data. For example, the model training system 150 can use the numerical values ​​of the parameters of at least one operation of a node in the second sample data as labels, compare them with the prediction results of the second model structure, and optimize the parameters of the second model structure through gradient updates or other methods, thereby improving its generalization ability and prediction accuracy in various business scenarios.

[0087] Return to reference Figure 4In box 434, the model training system 150 can evaluate the performance of the trained workflow creation model to determine whether its generated results meet the expected accuracy and reasonableness. For example, the model training system 150 can evaluate the model's performance in the workflow framework generation stage and the node configuration generation stage based on multiple creation requests in the test set. Evaluation methods may include, but are not limited to: structure matching accuracy, parameter configuration correctness, and process logic consistency.

[0088] In some embodiments, the model training system 150 can determine generation biases or coverage blind spots in the model based on low-quality samples (bad cases) identified during the evaluation process. For example, if the model has incorrect node order in a specific type of creation request (such as a nested judgment process), or if invalid fields are frequently entered in the node parameter configuration, the model training system 150 can identify these failure types and record their triggering conditions.

[0089] Furthermore, the model training system 150 can re-enter the training data construction process based on low-quality samples identified during the evaluation process. For example, process 400 can be re-executed to box 401. The model training system 150 can re-analyze the semantic requirements for these low-quality samples and construct targeted training samples. This process may include supplementing new creation requests, manually correcting node configurations, adjusting prompt word templates, or introducing new sample fields. In this way, the model training system 150 can achieve closed-loop optimization of training data, thereby continuously iterating the quality of training samples to improve the accuracy and generalization ability of the workflow creation model under various business requests.

[0090] In summary, according to the embodiments of this disclosure, by constructing first and second sample data based on multiple creation requests and corresponding data object information, and dividing the workflow creation task into two stages—framework generation and node configuration—the model's ability to express and understand complex process construction tasks and its ability to generate structures can be effectively improved. Furthermore, by adopting a phased training strategy to train the corresponding model structures separately, the workflow creation model can not only automatically construct the overall framework of the process but also accurately configure the parameters required for each node. In this way, the adaptability and creation quality of the workflow creation model under multiple business scenarios can be improved.

[0091] Figure 5 A flowchart of a method 500 for creating a model for a training workflow according to some embodiments of the present disclosure is shown. Method 500 can be implemented in any device. For example, method 500 can be implemented at a model training system 150, as referenced below. Figure 1 Description method 500.

[0092] In box 510, the model training system 150 receives a creation request for creating a workflow for a data object.

[0093] In box 520, the model training system 150 acquires data for the workflow corresponding to the creation request, which is created based on the creation request and information about the data object.

[0094] In box 530, the model training system 150 determines, based on the workflow data, first sample data related to the workflow framework and second sample data related to the configuration information of multiple nodes in the workflow framework.

[0095] In box 540, the model training system 150 creates a model based on the first sample data and the second sample data, using a training workflow.

[0096] In some embodiments, obtaining a creation request includes: obtaining requirement information associated with the creation workflow; and generating a creation request based on the requirement information.

[0097] In some embodiments, generating a creation request further includes: determining whether the creation request meets the quality requirements associated with the creation workflow; and adjusting the creation request in response to the creation request not meeting the quality requirements.

[0098] In some embodiments, the workflow includes a first workflow created via a graphical interface based on information of a creation request and a data object, and wherein obtaining data of the workflow corresponding to the creation request includes: determining whether the first workflow matches the creation request; and in response to the first workflow matching the creation request, extracting data of the first workflow according to a predetermined data format.

[0099] In some embodiments, the workflow further includes a second workflow recreated via a graphical interface based on information from the creation request and the data object, and the method further includes: in response to a mismatch between the first workflow and the creation request, extracting data from the second workflow in a predetermined data format.

[0100] In some embodiments, determining the first sample data and the second sample data includes: determining the first sample data based on workflow data, the first sample data including the topology of a plurality of nodes, the type of nodes among the plurality of nodes, and at least one operation of the nodes; and determining the second sample data based on workflow data and at least one operation of the nodes, the second sample data including parameters of at least one operation of the nodes, and the numerical values ​​of the parameters of at least one operation.

[0101] In some embodiments, the second sample data includes view information associated with a data object, the view information being used to indicate at least one of the following: filtering criteria or sorting methods for information about the data object.

[0102] In some embodiments, the workflow creation model includes a first model structure and a second model structure, and training the workflow creation model includes: training the first model structure based on first sample data, the first model structure being used to generate a prediction of the workflow framework; and training the second model structure based on second sample data, the second model structure being used to generate configuration information of multiple nodes in the prediction of the workflow framework.

[0103] In some embodiments, training the first model structure includes: generating a prediction of the workflow framework by invoking the first model structure; and training the first model structure based on the prediction of the workflow framework and the first sample data.

[0104] In some embodiments, the prediction of the workflow framework for generating the workflow includes: generating a prediction of the workflow framework based on the creation request, information about the data object, and a framework-generated prompt word template.

[0105] In some embodiments, training the second model structure includes: generating configuration information of multiple nodes in the prediction of the workflow framework by invoking the second model structure; and training the second model structure based on the configuration information of the multiple nodes and second sample data.

[0106] In some embodiments, generating configuration information for multiple nodes in the prediction of the workflow framework includes: for a node among the multiple nodes, generating a prompt word template based at least on the node's description information and configuration, and generating the numerical value of a parameter for at least one operation of the node as at least a part of the node's configuration information.

[0107] Embodiments of this disclosure also provide corresponding apparatus for implementing the above methods or processes. Figure 6 A schematic structural block diagram of an apparatus 600 for training a workflow to create a model is shown according to some embodiments of the present disclosure. Apparatus 600 may be implemented in or included in model training system 150, for example. Various modules / components in apparatus 600 may be implemented by hardware, software, firmware, or any combination thereof.

[0108] like Figure 6 As shown, the apparatus 600 includes a creation request acquisition module 610, configured to acquire a creation request for creating a workflow based on a data object; a workflow data acquisition module 620, configured to acquire workflow data corresponding to the creation request, wherein the workflow is created based on the creation request and information of the data object; a sample data determination module 630, configured to determine, based on the workflow data, first sample data related to the workflow framework and second sample data related to the configuration information of multiple nodes in the workflow framework; and a training module 640, configured to train a workflow creation model based on the first and second sample data.

[0109] In some embodiments, the creation request acquisition module 610 is further configured to: acquire requirement information associated with the creation workflow; and generate a creation request based on the requirement information.

[0110] In some embodiments, the creation request acquisition module 610 is further configured to: determine whether the creation request meets the quality requirements associated with the creation workflow; and adjust the creation request in response to the creation request not meeting the quality requirements.

[0111] In some embodiments, the workflow includes a first workflow created via a graphical interface based on information from a creation request and a data object, and the workflow data acquisition module 620 is further configured to: determine whether the first workflow matches the creation request; and in response to the first workflow matching the creation request, extract data from the first workflow according to a predetermined data format.

[0112] In some embodiments, the workflow also includes a second workflow recreated via a graphical interface based on the creation request and information of the data object, and the workflow data acquisition module 620 is further configured to: in response to a mismatch between the first workflow creation request and the second workflow, extract data from the second workflow in a predetermined data format.

[0113] In some embodiments, the sample data determination module 630 is further configured to: determine first sample data based on workflow data, the first sample data including the topology of a plurality of nodes, the type of nodes among the plurality of nodes, and at least one operation of the nodes; and determine second sample data based on workflow data and at least one operation of the nodes, the second sample data including parameters of at least one operation of the nodes, and the numerical values ​​of the parameters of at least one operation.

[0114] In some embodiments, the second sample data includes view information associated with a data object, the view information being used to indicate at least one of the following: filtering criteria or sorting methods for information about the data object.

[0115] In some embodiments, the workflow creation model includes a first model structure and a second model structure, and the training module 640 is further configured to: train the first model structure based on the first sample data, the first model structure being used to generate a prediction of the workflow framework; and train the second model structure based on the second sample data, the second model structure being used to generate configuration information of multiple nodes in the prediction of the workflow framework.

[0116] In some embodiments, the training module 640 is further configured to: generate a prediction of the workflow framework by invoking the first model structure; and train the first model structure based on the prediction of the workflow framework and the first sample data.

[0117] In some embodiments, the training module 640 is further configured to generate predictions for the workflow framework based on the creation request, information about the data object, and the framework to generate prompt word templates.

[0118] In some embodiments, the training module 640 is further configured to: generate configuration information of multiple nodes in the prediction of the workflow framework by invoking the second model structure; and train the second model structure based on the configuration information of the multiple nodes and the second sample data.

[0119] In some embodiments, the training module 640 is further configured to: generate, for a node among a plurality of nodes, at least based on the node's description information and configuration, generate a prompt word template, and generate a numerical value of at least one parameter of the node's operation as at least a part of the node's configuration information.

[0120] The units and / or modules included in device 600 can be implemented in various ways, including software, hardware, firmware, or any combination thereof. In some embodiments, one or more units and / or modules can be implemented using software and / or firmware, such as machine-executable instructions stored on a storage medium. In addition to or as an alternative to machine-executable instructions, some or all of the units and / or modules in device 600 can be implemented at least partially by one or more hardware logic components. By way of example and not limitation, exemplary types of hardware logic components that can be used include field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0121] Figure 7 A block diagram is shown of an electronic device 700 in which one or more embodiments of the present disclosure may be implemented. The electronic device 700 may, for example, be used to implement... Figure 1 The model training system shown is 150. It should be understood that... Figure 7 The electronic device 700 shown is merely exemplary and should not be construed as limiting the functionality and scope of the embodiments described herein.

[0122] refer to Figure 7The electronic device 700 is in the form of a general-purpose electronic device. Components of the electronic device 700 may include, but are not limited to, one or more processors or processor 710, memory 720, storage device 730, one or more communication units 740, one or more input devices 750, and one or more output devices 760. The processor 710 may be a physical or virtual processor and is capable of performing various processes according to programs stored in memory 720. In a multiprocessor system, multiple processing units execute computer-executable instructions in parallel to improve the parallel processing capability of the electronic device 700.

[0123] Electronic device 700 typically includes multiple computer storage media. Such media can be any available media accessible to electronic device 700, including but not limited to volatile and non-volatile media, removable and non-removable media. Memory 720 can be volatile memory (e.g., registers, cache, random access memory (RAM)), non-volatile memory (e.g., read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory), or some combination thereof. Storage device 730 can be removable or non-removable media and can include machine-readable media, such as flash drives, disks, or any other media capable of storing information and / or data and accessible within electronic device 700.

[0124] Electronic device 700 may further include additional removable / non-removable, volatile / non-volatile storage media. Although not explicitly stated... Figure 7 As shown, disk drives for reading from or writing to removable, non-volatile disks (e.g., "floppy disks") and optical disk drives for reading from or writing to removable, non-volatile optical disks can be provided. In these cases, each drive can be connected to a bus (not shown) via one or more data media interfaces. Memory 720 may include computer program product 725 having one or more program modules configured to perform various methods or actions of various embodiments of this disclosure.

[0125] The communication unit 740 enables communication with other electronic devices via a communication medium. Additionally, the functionality of the components of the electronic device 700 can be implemented using a single computing cluster or multiple computing machines capable of communicating via communication connections. Therefore, the electronic device 700 can operate in a networked environment using logical connections to one or more other servers, network personal computers (PCs), or another network node.

[0126] Input device 750 can be one or more input devices, such as a mouse, keyboard, trackball, etc. Output device 760 can be one or more output devices, such as a monitor, speaker, printer, etc. Electronic device 700 can also communicate with one or more external devices (not shown) via communication unit 740 as needed. These external devices include storage devices, display devices, etc., and can communicate with one or more devices that enable user interaction with electronic device 700, or with any device that enables electronic device 700 to communicate with one or more other electronic devices (e.g., network card, modem, etc.). Such communication can be performed via input / output (I / O) interface (not shown).

[0127] According to an exemplary implementation of this disclosure, a computer-readable storage medium is provided that stores computer-executable instructions thereon, wherein the computer-executable instructions are executed by a processor to implement the methods described above. According to an exemplary implementation of this disclosure, a computer program product is also provided, which is tangibly stored on a non-transitory computer-readable medium and includes computer-executable instructions, which are executed by a processor to implement the methods described above.

[0128] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatuses, devices, and computer program products implemented according to this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0129] These computer-readable program instructions can be provided to a processing unit of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processing unit of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner. Thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0130] Computer-readable program instructions can be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions that execute on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0131] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction, which contains one or more executable instructions for implementing the specified logical function. In some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0132] Various implementations of this disclosure have been described above. The foregoing description is exemplary and not exhaustive, nor is it limited to the disclosed implementations. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described implementations. The terminology used herein is determined to best explain the principles, practical applications, or improvements to technology in the market, or to enable others skilled in the art to understand the various implementations disclosed herein.

Claims

1. A method for training a workflow creation model, comprising: Retrieve the creation request for creating a workflow based on a data object; Obtain the workflow data corresponding to the creation request, wherein the workflow is created based on the creation request and the information of the data object; Based on the data of the workflow, determine first sample data related to the workflow framework and second sample data related to the configuration information of multiple nodes in the workflow framework; as well as The workflow creation model is trained based on the first sample data and the second sample data.

2. The method according to claim 1, wherein obtaining the creation request includes: Obtain the requirements information associated with creating the workflow; as well as Based on the aforementioned requirement information, the creation request is generated.

3. The method according to claim 2, wherein generating the creation request further comprises: Determine whether the creation request meets the quality requirements associated with creating the workflow; as well as In response to the creation request not meeting the quality requirements, the creation request is adjusted.

4. The method of claim 1, wherein the workflow includes a first workflow created via a graphical interface based on the creation request and information of the data object, and wherein obtaining data for the workflow corresponding to the creation request includes: Determine whether the first workflow matches the creation request; as well as In response to the creation request matching the first workflow, data from the first workflow is extracted according to a predetermined data format.

5. The method of claim 4, wherein the workflow further comprises a second workflow recreated via the graphical interface based on the creation request and the information of the data object, and the method further comprises: In response to the first workflow not matching the creation request, data from the second workflow is extracted according to the predetermined data format.

6. The method of claim 1, wherein determining the first sample data and the second sample data comprises: Based on the workflow data, the first sample data is determined, and the first sample data includes the topology of the plurality of nodes, the type of the nodes among the plurality of nodes, and at least one operation of the nodes; as well as Based on the workflow data and the at least one operation of the node, the second sample data is determined, the second sample data including the parameters of the at least one operation of the node and the values ​​of the parameters of the at least one operation.

7. The method of claim 1, wherein the second sample data includes view information associated with the data object, the view information indicating at least one of the following: filtering conditions or sorting methods for the information of the data object.

8. The method according to claim 1, wherein the workflow creation model includes a first model structure and a second model structure, and training the workflow creation model includes: The first model structure is trained based on the first sample data, and the first model structure is used to generate a prediction of the workflow framework of the workflow. as well as The second model structure is trained based on the second sample data, and the second model structure is used to generate configuration information for multiple nodes in the prediction of the workflow framework.

9. The method of claim 8, wherein training the first model structure comprises: The workflow framework prediction is generated by calling the first model structure; as well as The first model structure is trained based on the predictions made by the workflow framework and the first sample data.

10. The method of claim 9, wherein the prediction of the workflow framework that generates the workflow comprises: Based on the creation request, the information of the data object, and the framework, a prompt word template is generated to produce the prediction of the workflow framework.

11. The method of claim 8, wherein training the second model structure comprises: By invoking the second model structure, configuration information for multiple nodes in the prediction of the workflow framework is generated; as well as The second model structure is trained based on the configuration information of the multiple nodes and the second sample data.

12. The method of claim 11, wherein generating configuration information for the plurality of nodes in the prediction of the workflow framework includes: For each of the plurality of nodes, a prompt word template is generated based at least on the node's description information and configuration, and a parameter value for at least one operation of the node is generated as at least a part of the node's configuration information.

13. An apparatus for training a workflow creation model, comprising: The request retrieval module is configured to retrieve creation requests for creating workflows based on data objects. The workflow data acquisition module is configured to acquire data of the workflow corresponding to the creation request, wherein the workflow is created based on the creation request and the information of the data object; The sample data determination module is configured to determine, based on the data of the workflow, first sample data related to the workflow framework and second sample data related to the configuration information of multiple nodes in the workflow framework; as well as The training module is configured to train the workflow creation model based on the first sample data and the second sample data.

14. An electronic device, comprising: At least one processor; as well as At least one memory coupled to the at least one processor and storing instructions for execution by the at least one processor, the instructions causing the electronic device to perform the method according to any one of claims 1 to 12 when executed by the at least one processor.

15. A computer-readable storage medium having stored thereon computer-executable instructions that can be executed by a processor to implement the method according to any one of claims 1 to 12.

16. A computer program product comprising computer-executable instructions, wherein the computer-executable instructions, when executed by a processor, implement the method according to any one of claims 1 to 12.

Images