AI workflow automation management method based on creative scenarios

Abstract

The present invention provides an AI workflow automation management method based on creative scenarios, and relates to the technical field of data processing. The method comprises: obtaining initial data input by a user, extracting key elements from text data, analyzing relevant characteristics from non-text reference data, constructing an initial task chain, analyzing dependencies of task units, generating a dependency path set, weighting dependencies between task units in the dependency path set, optimizing the initial task chain, generating a dynamic task chain table, selecting applicable tools from an AI tool library, and allocating corresponding tool resources to each task unit, constructing an execution workflow, dynamically running the execution workflow, obtaining intermediate result data, detecting whether the intermediate result data meets requirements of logical relationships and content styles, adjusting inconsistent parts, and generating output result data. The present invention improves the intelligence level of the workflow.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to an AI workflow automation management method based on creative scenarios. Background Art

[0002] In recent years, the application of artificial intelligence technology in creative scenarios has gradually expanded, especially in the fields of content generation, artistic creation, and marketing planning, with the help of AI tools to achieve partial automation of workflows. Existing technologies usually use predefined workflow rules and model algorithms, combined with image generation, text generation and other technologies, to complete the generation of creative content that meets user needs. For example, image generation technology based on deep learning can generate pictures with a specific style based on input keywords, and text generation technology based on natural language processing can generate copywriting content related to specified topics. However, in traditional implementation methods, different AI tools are highly independent, and it is difficult to form efficient collaboration across tools. Users need to manually switch between multiple tools and integrate the results, which increases the complexity of operations.

[0003] However, existing technologies have obvious technical defects in cross-tool collaborative application scenarios. For example, when generating a comprehensive creative plan for a marketing campaign, users need to call multiple AI tools in sequence to generate copy, design drawings, and schedules respectively, and integrate them manually. Due to the lack of a unified management method, the style and logic of the results generated by different tools may be inconsistent, resulting in the need for further manual adjustment of the generated plan to ensure the coherence and consistency of the content. This manual adjustment process is not only complicated, but may also reduce the objectivity and quality of the results due to the subjective intervention of the user, making it difficult to meet large-scale automation needs. Summary of the invention

[0004] The purpose of the present invention is to provide an AI workflow automation management method based on creative scenarios, aiming to solve the problems mentioned in the background technology.

[0005] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0006] An AI workflow automation management method based on creative scenarios, the method comprising:

[0007] Acquiring initial data input by a user, wherein the initial data includes text data describing a target creative scene and non-text reference data;

[0008] Based on the initial data, key elements are extracted from the text data, and relevant features are analyzed from the non-text reference data to obtain parsed data;

[0009] Constructing an initial task chain according to the parsed data, wherein the initial task chain includes a plurality of task units and their logical association paths;

[0010] According to the logical association path, the dependency relationship of the task unit is analyzed to generate a set of dependency paths. The dependency path is used to describe the direct data transfer and execution order between the task units.

[0011] The dependency relationships between task units in the dependency path set are weighted and priorities are assigned to the results to generate a dynamic path table, which includes the execution order of task units, the transmission order of dependency data, and adjustable backup paths.

[0012] Optimize the initial task chain according to the dynamic path table and generate a dynamic task chain table;

[0013] According to the execution requirements of the task units in the dynamic task list, select the appropriate tools from the AI ​​tool library, allocate corresponding tool resources to each task unit, and generate matching data;

[0014] Constructing an execution workflow according to the matching data, wherein the execution workflow includes the execution order and data transfer path of the task units, and associating the execution rules of each task unit with the corresponding tool;

[0015] Dynamically run the execution workflow to obtain intermediate result data, wherein the intermediate result data includes the content generated by each task unit;

[0016] Check whether the intermediate result data meets the requirements of logical relationship and content style, adjust the inconsistent parts, and generate task unit optimization data;

[0017] The task unit optimization data is classified, formatted and consistency-checked according to the target requirements of the execution workflow to generate output result data, which includes text, images or multimedia content generated according to the user's target requirements.

[0018] Preferably, based on the initial data, key elements are extracted from the description information, and relevant characteristics are analyzed from the reference content to obtain parsed data, including:

[0019] According to the text data, the text sentences are decomposed, the keywords are located, and the first content data is extracted by analyzing the context relevance;

[0020] By content segmentation, local correlation characteristics of non-text reference data are marked and classified, feature extraction is performed, and second content data is generated;

[0021] grouping the first content data and the second content data according to category characteristics to generate characteristic grouping data;

[0022] Group data according to each feature, extract correlation features based on the cross-matching method, and build a multi-layer data association network;

[0023] Filter subset data that meet the preset correlation threshold from the multi-layer data association network to generate parsed data.

[0024] Preferably, an initial task chain is constructed based on the parsed data, including:

[0025] Extracting information elements of the task unit from the parsed data, the information elements including task objectives, input data formats, and expected output types;

[0026] Analyze information elements and build logical association paths based on the dependencies between task objectives;

[0027] Use logical association paths to mark the data flow and execution order between task units and generate task dependencies;

[0028] According to the task dependency, the task units are arranged in sequence according to the logical association path to build the initial task chain;

[0029] Assign a unique identifier to each task unit in the initial task chain to record its task objectives, input data sources, and output target destinations.

[0030] Preferably, according to the logical association path, the dependency relationship of the task unit is analyzed to generate a dependency path set, including:

[0031] Extract the logical association path of each task unit from the initial task chain, and establish a direct dependency path based on the data input and output relationship between task units;

[0032] Classify the upstream and downstream task units of each direct dependency path to form a set of strong dependency paths and weak dependency paths;

[0033] Analyze the interaction frequency of task units in the strong dependency path set and calculate the dependency strength between task units;

[0034] According to the dependency strength, priorities are assigned to the task units in the strong dependency path set, and a dependency path set is generated. The dependency path set includes direct data transfer and execution order between task units.

[0035] Preferably, the dependency relationships between the task units in the dependency path set are weighted and priorities are assigned to the results to generate a dynamic path table, including:

[0036] Obtain the task unit information of each path from the dependent path set, including the input data source, output target destination and dependency strength of the task unit;

[0037] According to the data flow frequency, path length and data processing format of the task unit, the dependency relationship between the task units in the dependency path set is weighted to obtain the dependency weight value; the calculation formula of the dependency weight value is as follows:

[0038] ,

[0039] in, Task Unit With task unit The dependency weight value between Task Unit With task unit The frequency of data flow between Task Unit With task unit The path length between Task Unit With task unit The format matching between Task Unit With task unit The path abnormality rate between , indicates the probability of interruption or delay of the path, Task Unit With task unit The conflict rate between them indicates the frequency of task resource conflicts. , , , , and is the weight coefficient;

[0040] The dependency weight values ​​are divided according to the preset dependency thresholds to generate a dependency path priority table, which is used to optimize the execution order of the task units;

[0041] According to the dependency path priority table, the execution order of the task units is arranged to generate a dynamic path table, which includes the execution order of the task units, the transmission path of the dependency data and the backup path information.

[0042] Preferably, the initial task chain is optimized according to the dynamic path table to generate a dynamic task chain table, including:

[0043] Extracting task unit information from the dynamic path table, including the execution order of the task units, the transmission path of dependent data, and the backup path information;

[0044] According to the transmission path of dependent data, the logical association path in the initial task chain is adjusted to form an optimized task path;

[0045] Allocate redundant execution paths for task units in the initial task chain according to the backup path information;

[0046] Integrate the optimized task path and redundant execution path, rearrange the execution order of task units in the initial task chain, and generate a dynamic task chain table;

[0047] According to the content of the dynamic task linked list, the execution rules and path information of each task unit are updated to the task dependency relationship, forming a dependency record including the execution order of the task units, dependency weight values ​​and path redundancy schemes.

[0048] Preferably, according to the execution requirements of the task units in the dynamic task list, an applicable tool is selected from the AI ​​tool library, and corresponding tool resources are allocated to each task unit to generate matching data, including:

[0049] Extract the execution requirements of task units from the dynamic task list, including input data format, output target and processing time limit;

[0050] According to the execution requirements of the task unit, the tool resource with the highest adaptability is allocated from the AI ​​tool library, and the tool allocation information is generated. The tool allocation information includes the task unit identifier, tool identifier, and tool adaptation parameters. The calculation formula of the adaptability is as follows:

[0051] ,

[0052] in, Task Unit The adaptability, For tools The adaptability of the tool Satisfy the task unit The ability to execute requirements, For tools The average processing time, Task Unit The upper limit of the processing time indicates the time constraint of the task. Task Unit The execution priority reflects the priority of the task unit in the dynamic task list. Task Unit The resource demand degree indicates the number of tool resources that the task unit needs to call. Task Unit The tool switching cost represents the time or resource cost of switching from other tools to the current tool. For tools The weight coefficient of and is the adjustment coefficient, is the number of tools;

[0053] Input the tool allocation information into the dynamic task chain table to generate the tool calling path of the task unit;

[0054] Matching data is formed according to the tool allocation information and the tool calling path.

[0055] Preferably, the execution workflow is constructed according to the matching data, including:

[0056] Extract the execution order and tool calling path of each task unit from the dynamic task linked list;

[0057] According to the execution order of task units, establish the data transmission path between task units;

[0058] The execution workflow is constructed based on the tool calling path and data transmission path. The execution workflow includes the execution rules of the task unit, data flow and tool calling relationship.

[0059] Preferably, dynamically running the execution workflow to obtain intermediate result data includes:

[0060] Run the task units in the execution order of the task units in the execution workflow, and call the corresponding tool resources to process the input data during the running process;

[0061] After each task unit completes execution, it generates intermediate result data and transmits the result to the subsequent task unit through the data transmission path;

[0062] Record the operation log of each task unit, including input and output data and execution time.

[0063] Preferably, detecting whether the intermediate result data meets the requirements of logical relationship and content style, and adjusting the inconsistent parts to generate task unit optimization data includes:

[0064] Verify the format, logic and content style of each intermediate result data, and obtain the data that failed the verification and the task units to be adjusted;

[0065] According to the data that failed the verification, the execution rules of the task unit to be adjusted or the tool adaptation parameter problem are analyzed to obtain the analysis result data;

[0066] According to the analysis result data, the execution rule parameters of the task unit to be adjusted are adjusted or the tools are reallocated, and the task unit to be adjusted is re-run to generate task unit optimization data.

[0067] The above solution of the present invention includes at least the following beneficial effects:

[0068] By integrating text data and non-text reference data, a dynamic and flexible task chain optimization mechanism is constructed, which fundamentally solves the problem of insufficient cross-tool collaboration in the existing technology. Compared with the traditional single tool or independent tool mode, this invention parses the multimodal data input by the user to form a unified parsed data, laying a solid foundation for workflow automation.

[0069] In traditional technologies, users need to manually switch between multiple tools and integrate the results, which causes operational complexity and consistency issues. However, this invention systematically organizes different task units and their logical association paths by constructing an initial task chain, ensuring that the dependencies between tasks are clear and efficient. Through the analysis of the logical association paths, the generated dependency path set can accurately describe the direct data transfer and execution order between task units, avoiding errors and incoherence in manual operations.

[0070] The dependencies between task units in the dependent path set are converted into a dynamic path table with optimization capabilities through weight calculation and priority allocation. The dynamic path table not only contains the execution order of task units, but also includes backup path information, thereby ensuring the continuous execution of tasks in abnormal situations. This dynamic adjustment capability solves the pain point of inconsistent generation results in existing technologies and provides reliable protection for large-scale automated workflows.

[0071] In addition, efficient cross-tool collaboration is achieved by automatically adapting the task unit execution requirements to the AI ​​tool library. Unlike the traditional method that requires users to manually assign tools, the present invention can dynamically select the optimal tool according to task requirements and allocate appropriate tool resources to each task unit. This fully automated resource allocation method not only improves operational efficiency, but also makes the content generated by different tools more consistent in style and logic, thereby avoiding manual adjustments caused by mismatched results of multiple tools in traditional methods.

[0072] For the final generated content, the logic and consistency of the generated results are further improved by dynamically running the execution workflow and real-time detection of intermediate results. In the prior art, due to the lack of a unified management framework, the content generated by different tools needs to be manually adjusted in style and logic. The present invention realizes automatic adjustment through logical verification and style consistency detection of intermediate results, ensuring that the output results can directly meet user needs and reducing manual intervention.

[0073] In summary, the present invention significantly reduces the complexity of users switching between multiple AI tools, enhances the cross-tool task collaboration capability, and improves the automation level of the task chain and the quality of generated content through dynamic path table optimization and efficient adaptation of task units and tools. It not only meets the needs of efficiency and automation in creative scenarios, but also provides users with more consistent and coherent creative solutions, fundamentally overcoming the limitations of existing technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 It is a flowchart of an AI workflow automation management method based on creative scenarios provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0075] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0076] like Figure 1 As shown, an embodiment of the present invention proposes an AI workflow automation management method based on a creative scenario, and the method includes:

[0077] S100, obtaining initial data input by a user, wherein the initial data includes text data describing a target creative scene and non-text reference data;

[0078] S200, extracting key elements from text data based on initial data, and analyzing relevant characteristics from non-text reference data to obtain parsed data;

[0079] S300, constructing an initial task chain according to the parsed data, wherein the initial task chain includes a plurality of task units and their logical association paths;

[0080] S400, analyzing the dependency relationship of the task units according to the logical association path, and generating a dependency path set, where the dependency path is used to describe the direct data transfer and execution order between the task units;

[0081] S500, weighting the dependency relationships between the task units in the dependency path set, assigning priorities to the results, and generating a dynamic path table, which includes the execution order of the task units, the transmission order of the dependency data, and an adjustable backup path;

[0082] S600, optimizing the initial task chain according to the dynamic path table to generate a dynamic task chain table;

[0083] S700, according to the execution requirements of the task units in the dynamic task chain table, select applicable tools from the AI ​​tool library, allocate corresponding tool resources to each task unit, and generate matching data;

[0084] S800, constructing an execution workflow according to the matching data, wherein the execution workflow includes the execution order and data transfer path of the task units, and associating the execution rules of each task unit with the corresponding tool;

[0085] S900, dynamically run the execution workflow to obtain intermediate result data, where the intermediate result data includes the content generated by each task unit;

[0086] S1000, checking whether the intermediate result data meets the requirements of logical relationship and content style, and adjusting the inconsistent parts to generate task unit optimization data;

[0087] S1100, classify, format and check consistency of task unit optimization data according to the target requirements of executing the workflow, and generate output result data, which includes text, images or multimedia content generated according to the user's target requirements.

[0088] In an embodiment of the present invention, by obtaining the initial data input by the user, the characteristics of text data and non-text reference data are comprehensively processed, providing a highly adaptable solution for data analysis that describes the target creative scene. First, the processing of text data ensures the accurate extraction of key elements, while non-text reference data provides additional background support for the construction of task chains through the analysis of relevant characteristics. This multimodal data processing method enables the method to cover multiple types of inputs and adapt to the complex and diverse needs of creative scenes. In addition, the generation of parsed data realizes the semantic connection between data, laying the foundation for the logical planning of subsequent task chains.

[0089] The construction of the initial task chain greatly simplifies the design of complex task processes. It contains multiple task units and their logically associated paths, so that the dependencies between task units can be optimized through the dynamic path table. The generation of the dependency path set helps to achieve an accurate description of the direct data transfer and execution order between task units, providing support for the efficient execution of the task chain. At the same time, through the priority sorting of the dynamic path table, not only the flexibility of the data transfer path is significantly improved, but also the backup path can ensure the continuous operation of the task in abnormal situations, enhancing the robustness of the overall method.

[0090] By adapting the task execution requirements to the AI ​​tool library through the dynamic task chain table, this method can dynamically select tool resources in different scenarios, thereby maximizing resource utilization efficiency in the process of building the execution workflow. The dynamic operation of the execution workflow further enhances the automation and adaptability of the task. Each task unit can provide instant feedback and adjustments after generating intermediate result data, thereby ensuring the logical consistency of the final output result and the high quality of the content style. In short, this method not only effectively solves the construction and optimization problems of the task chain in creative scenarios, but also greatly improves the efficiency of automated execution and the quality of results.

[0091] More specifically, the task unit optimization data is classified, formatted, and consistency-checked according to the target requirements of the execution workflow to generate output result data, which includes text, images, or multimedia content generated according to the user's target requirements. The specific steps include:

[0092] Classification: Grouping task unit optimization data according to the goal requirements of executing the workflow. For example, if the user's goal includes generating a marketing plan, there may be text data such as copywriting, image data such as product design drawings, and multimedia data such as promotional videos. The core basis for classification is the target use and attribute characteristics of the data, such as format, content type, and usage scenario.

[0093] Formatting: refers to adjusting the classified data according to specific standards so that the data meets user needs in terms of presentation. Specific operations include: adjusting fonts, paragraph structure and content order to make the copy clearer. Adjusting resolution, size and color style to match the output target. Optimizing encoding format, resolution and audio and video synchronization according to needs.

[0094] Consistency check: Verify the consistency of data in logic, style, and semantics. For example, whether the content between text and images is coherent. Whether the keywords mentioned in the text are consistent with the image or multimedia content. Whether the data meets the specific style requirements provided by the user, such as brand style or artistic style.

[0095] Output result data: Output result data generated after classification, formatting and consistency verification includes copywriting content generated according to user needs, design drawings of a specific style, and complete promotional materials including audio and video content.

[0096] In a preferred embodiment of the present invention, according to the initial data, key elements are extracted from the description information, and relevant characteristics are analyzed from the reference content to obtain parsed data, including:

[0097] According to the text data, the text sentences are decomposed, the keywords are located, and the first content data is extracted by analyzing the context relevance;

[0098] By content segmentation, local correlation characteristics of non-text reference data are marked and classified, feature extraction is performed, and second content data is generated;

[0099] grouping the first content data and the second content data according to category characteristics to generate characteristic grouping data;

[0100] Group data according to each feature, extract correlation features based on the cross-matching method, and build a multi-layer data association network;

[0101] Filter subset data that meet the preset correlation threshold from the multi-layer data association network to generate parsed data.

[0102] In an embodiment of the present invention, by extracting key elements from the descriptive information and analyzing the relevant characteristics of the reference content, the method realizes fine-grained processing of input data, especially in multimodal data fusion. The decomposition of text data and keyword positioning ensure the accuracy of semantic extraction, while the feature annotation and classification of non-text reference data provide technical support for the structuring of complex data. The generation of feature grouping data integrates heterogeneous data into a unified framework, further extracts correlation features through cross-matching methods, and constructs a multi-layer data association network. This network structure effectively captures the implicit associations between data, forms high-quality parsed data, and provides a strong guarantee for the subsequent task chain design.

[0103] More specifically, according to the text data, the text sentences are decomposed, the keywords are located, and the first content data is extracted by analyzing the context relevance. The specific steps include:

[0104] Sentence decomposition: Decompose the long text entered by the user into independent sentence units. For example, divide a paragraph of text describing brand positioning into multiple short sentences, each of which independently describes a feature or requirement.

[0105] Keyword positioning: extract core keywords from each sentence, such as "target market", "consumer preferences", etc. Keyword extraction can be achieved through word frequency analysis, named entity recognition (NER) or context-dependent models.

[0106] Analyze context relevance: Through natural language processing algorithms such as bidirectional language models, analyze the relevance between keywords and context, extract the most relevant information to the target scenario, and form the first content data. The first content data is a set of structured text fragments that contain the core information input by the user.

[0107] More specifically, the local correlation characteristics of the non-text reference data are annotated and classified through content segmentation, and feature extraction is performed to generate the second content data. The specific steps include:

[0108] Content segmentation: Local segmentation of input non-text data, such as segmenting an image into different regions or cutting a video into multiple key frames.

[0109] Local correlation feature annotation and classification: Annotate each segment with features related to the user’s goal, such as color or texture features of images or action categories of videos.

[0110] Feature extraction: Extract feature vectors of non-text data through deep learning models such as convolutional neural networks to generate secondary content data. The secondary content data is structured data after classification and feature extraction, which is convenient for subsequent processing.

[0111] More specifically, the first content data and the second content data are grouped according to category characteristics to generate characteristic grouping data, and the specific steps include:

[0112] The first content data and the second content data are clustered according to category characteristics such as theme, style, target scene, etc.

[0113] Each group of data contains data items with the same characteristics, for example, all texts and images related to "young consumers" are grouped together.

[0114] More specifically, data is grouped according to each feature, correlation features are extracted based on the cross-matching method, and a multi-layer data association network is constructed. The specific steps include:

[0115] Cross-matching method: compare text data and non-text data to extract relevant features. For example, the "fresh style" described in the text can be verified by the "bright tones" in the image features.

[0116] Multi-layer data association network: With association features as edges and data groups as nodes, a graph-structured association network is constructed. The layers of the multi-layer network may represent different feature dimensions, such as "theme layer", "style layer", etc.

[0117] More specifically, subset data that meets a preset correlation threshold is selected from the multi-layer data association network to generate parsed data. The specific steps include:

[0118] Preset correlation threshold: A threshold set according to user needs, used to filter data pairs with high correlation. For example, only data pairs with a correlation higher than 80% can be retained.

[0119] Generate parsed data: The final parsed data is a filtered subset that contains the most relevant information for the user's target scenario. These parsed data provide reliable input for the construction of the task chain.

[0120] In a preferred embodiment of the present invention, an initial task chain is constructed based on the parsed data, including:

[0121] Extracting information elements of the task unit from the parsed data, the information elements including task objectives, input data formats, and expected output types;

[0122] Analyze information elements and build logical association paths based on the dependencies between task objectives;

[0123] Use logical association paths to mark the data flow and execution order between task units and generate task dependencies;

[0124] According to the task dependency, the task units are arranged in sequence according to the logical association path to build the initial task chain;

[0125] Assign a unique identifier to each task unit in the initial task chain to record its task objectives, input data sources, and output target destinations.

[0126] In an embodiment of the present invention, parsed data is used to construct an initial task chain, a process that fundamentally optimizes the logical associations between task units. The extraction of task unit information elements, including task objectives, input data formats, and expected output types, ensures that the definition of each task unit is clear and unambiguous. The generation of logical association paths lays a foundation for the accuracy of the relationship between task units, making the data flow and execution order traceable and reasonable. Task dependencies significantly improve the standardization level of the initial task chain by further structuring the logical association paths. Independently identified task unit information provides a complete execution record, greatly facilitating subsequent task scheduling and result verification.

[0127] More specifically, the task dependency is a structure that describes the logical relationship between task units and is used to clarify the input and output sequence and data transfer relationship of the task units.

[0128] In a preferred embodiment of the present invention, the dependency relationship of the task unit is analyzed according to the logical association path to generate a dependency path set, including:

[0129] Extract the logical association path of each task unit from the initial task chain, and establish a direct dependency path based on the data input and output relationship between task units;

[0130] Classify the upstream and downstream task units of each direct dependency path to form a set of strong dependency paths and weak dependency paths;

[0131] Analyze the interaction frequency of task units in the strong dependency path set and calculate the dependency strength between task units;

[0132] According to the dependency strength, priorities are assigned to the task units in the strong dependency path set, and a dependency path set is generated. The dependency path set includes direct data transfer and execution order between task units.

[0133] In an embodiment of the present invention, by analyzing the logically associated paths, the method effectively generates a set of dependent paths, which not only marks the direct data transfer and execution order between task units, but also forms a set of strong dependent paths and weak dependent paths through classification, thereby improving the precision of path analysis. The calculation of the interaction frequency of task units in the strong dependent path set provides a quantitative basis for the reasonable allocation of task priorities. The path set significantly improves the scalability and adaptability of the system in optimizing the execution order of task units and the flexibility of data transfer, and provides support for the efficient operation of complex task chains.

[0134] More specifically, the upstream and downstream task units of each direct dependency path are classified to form a set of strong dependency paths and weak dependency paths, including:

[0135] A direct dependency path refers to a path where there is a direct data transfer or execution order relationship between task units. For example, the output of task unit A is directly used as the input of task unit B. This relationship constitutes a direct dependency path. Upstream and downstream task units refer to the source node (upstream) and the target node (downstream) in the path. The main basis for classification is the degree of dependency between task units, that is, whether the dependency is critical to the execution of subsequent tasks.

[0136] Strong dependency path: If the downstream task unit is highly dependent on the output of the upstream task unit and cannot be replaced by other paths, the path is considered a strong dependency path.

[0137] Weak dependency path: If the dependency of a downstream task unit on an upstream task unit can be replaced by other paths or redundant means, the path is considered a weak dependency path.

[0138] Forming path sets: Based on the classification of direct dependency paths, two sets are generated. The strong dependency path set contains all strong dependency paths and is used to describe the core dependency relationship of the task unit. The weak dependency path set contains all weak dependency paths and is used to assist in analyzing the redundant relationship of the task unit or the design of alternate paths.

[0139] In a preferred embodiment of the present invention, the dependency relationships between the task units in the dependency path set are weighted and priorities are assigned to the results to generate a dynamic path table, including:

[0140] Obtain the task unit information of each path from the dependent path set, including the input data source, output target destination and dependency strength of the task unit;

[0141] According to the data flow frequency, path length and data processing format of the task unit, the dependency relationship between the task units in the dependency path set is weighted to obtain the dependency weight value; the calculation formula of the dependency weight value is as follows:

[0142] ,

[0143] in, Task Unit With task unit The dependency weight between Task Unit With task unit The frequency of data flow between Task Unit With task unit The path length between Task Unit With task unit The format matching between Task Unit With task unit The path abnormality rate between , indicates the probability of interruption or delay of the path, Task Unit With task unit The conflict rate between them indicates the frequency of task resource conflicts. , , , , and is the weight coefficient;

[0144] The dependency weight values ​​are divided according to the preset dependency thresholds to generate a dependency path priority table, which is used to optimize the execution order of the task units;

[0145] According to the dependency path priority table, the execution order of the task units is arranged to generate a dynamic path table, which includes the execution order of the task units, the transmission path of the dependency data and the backup path information.

[0146] In an embodiment of the present invention, the dependencies between task units in the dependent path set are accurately quantified through weight calculation, and the data flow frequency, path length and task importance are comprehensively considered, reflecting the multi-dimensional characteristics of the path weight. The generation of the dependent path priority table not only clearly shows the relative importance of the path, but also guides the optimization process of the dynamic path table. The dynamic path table realizes the dynamic adjustment capability of the task chain by clarifying the execution order of the task units, the dependent data transmission path and the backup path information, thereby greatly improving the flexibility and stability of the task.

[0147] In a preferred embodiment of the present invention, the initial task chain is optimized according to the dynamic path table to generate a dynamic task chain table, including:

[0148] Extracting task unit information from the dynamic path table, including the execution order of the task units, the transmission path of dependent data, and the backup path information;

[0149] According to the transmission path of dependent data, the logical association path in the initial task chain is adjusted to form an optimized task path;

[0150] Allocate redundant execution paths for task units in the initial task chain according to the backup path information;

[0151] Integrate the optimized task path and redundant execution path, rearrange the execution order of task units in the initial task chain, and generate a dynamic task chain table;

[0152] According to the content of the dynamic task linked list, the execution rules and path information of each task unit are updated to the task dependency relationship, forming a dependency record including the execution order of the task units, dependency weight values ​​and path redundancy schemes.

[0153] In an embodiment of the present invention, the optimized design of the dynamic path table is used to adjust the initial task chain. By modifying the logically associated path and integrating the backup path, the optimized task path has a higher execution efficiency. The allocation of redundant execution paths ensures the continuity of the task chain under abnormal circumstances, and the rearranged execution order of the task units significantly improves the adaptability of the method to dynamic environments. The optimized dynamic task chain table, as a complete record of the task unit execution rules and path information, reflects the efficiency of the method in task chain management and dynamic updating.

[0154] More specifically, according to the transmission path of the dependent data, the logical association path in the initial task chain is adjusted to form an optimized task path. The specific steps include:

[0155] Refers to the actual paths for data flow between task units specified in the dynamic path table. These paths can be primary paths or include backup paths.

[0156] According to the transmission path of dependent data, the logical association path of the task unit is adjusted to make it more suitable for actual execution requirements. For example, when there are multiple association paths between task units, the efficient transmission path should be retained first.

[0157] The adjusted logical association paths are reorganized according to the dependency strength and data flow to generate optimized task paths. The optimized task paths eliminate redundant paths, reduce possible resource waste, and significantly improve the execution efficiency of the task chain.

[0158] More specifically, the optimized task path and the redundant execution path are integrated, the execution order of the task units in the initial task chain is rearranged, and a dynamic task chain table is generated. The specific steps include:

[0159] Optimize the task path: As the main path, it is used for task execution first.

[0160] Redundant execution path: A backup path used as an alternative when the main path fails.

[0161] Combine the two types of paths together to form a complete task chain structure. For each task unit, not only its main path information but also possible backup paths are recorded. According to the priority recorded in the dynamic path table, the execution order of the task units is rearranged according to the delivery order of dependent data. Dynamically switch between the main path and the redundant path to ensure the continuity of the task chain in abnormal situations.

[0162] The generation of dynamic task list includes:

[0163] Execution rules of task units: record the execution conditions and constraints of each task.

[0164] Path information: includes a complete description of the primary and backup paths.

[0165] Dependency record: the source and destination of dependency data between task units.

[0166] By generating a dynamic task chain list, the task chain has higher flexibility and robustness, and can adapt to dynamic adjustment needs in complex scenarios.

[0167] In a preferred embodiment of the present invention, according to the execution requirements of the task units in the dynamic task chain table, suitable tools are selected from the AI ​​tool library, and corresponding tool resources are allocated to each task unit to generate matching data, including:

[0168] Extract the execution requirements of task units from the dynamic task list, including input data format, output target and processing time limit;

[0169] According to the execution requirements of the task unit, the tool resource with the highest adaptability is allocated from the AI ​​tool library, and the tool allocation information is generated. The tool allocation information includes the task unit identifier, tool identifier, and tool adaptation parameters. The calculation formula of the adaptability is as follows:

[0170] ,

[0171] in, Task Unit The adaptability, For tools The adaptability of the tool Satisfy the task unit The ability to execute requirements, For tools The average processing time, Task Unit The upper limit of the processing time indicates the time constraint of the task. Task Unit The execution priority reflects the priority of the task unit in the dynamic task list. Task Unit The resource demand degree indicates the number of tool resources that the task unit needs to call. Task Unit The tool switching cost represents the time or resource cost of switching from other tools to the current tool. For tools The weight coefficient of and is the adjustment coefficient, is the number of tools;

[0172] Input the tool allocation information into the dynamic task chain table to generate the tool calling path of the task unit;

[0173] Matching data is formed according to the tool allocation information and the tool calling path.

[0174] In an embodiment of the present invention, the execution requirements of the task unit are extracted according to the dynamic task list, and the method realizes efficient matching between the AI ​​tool library and the task unit. By analyzing the input data format, output target and processing time limit, the allocation of tool resources realizes the maximum adaptability, and the design of the tool call path further optimizes the execution process of the task unit. The generation of tool allocation information not only provides a clear resource allocation plan for task execution, but also provides a solid foundation for the precise construction of execution workflow through the generation of matching data.

[0175] In a preferred embodiment of the present invention, the execution workflow is constructed according to the matching data, including:

[0176] Extract the execution order and tool calling path of each task unit from the dynamic task linked list;

[0177] According to the execution order of task units, establish the data transmission path between task units;

[0178] The execution workflow is constructed based on the tool calling path and data transmission path. The execution workflow includes the execution rules of the task unit, data flow and tool calling relationship.

[0179] In the embodiment of the present invention, matching data is used to construct an execution workflow, in which the combination of the execution order of task units and the tool call path significantly improves the efficiency of the data transfer path. The integration of execution rules and data flow ensures the coordinated operation of the task chain, and the introduction of tool call relationships provides technical support for the automated execution of complex tasks. The design of this workflow not only realizes the standardization and modularization of the task chain, but also facilitates future expansion and optimization.

[0180] In a preferred embodiment of the present invention, dynamically running the execution workflow to obtain intermediate result data includes:

[0181] Run the task units in the execution order of the task units in the execution workflow, and call the corresponding tool resources to process the input data during the running process;

[0182] After each task unit completes execution, it generates intermediate result data and transmits the result to the subsequent task unit through the data transmission path;

[0183] Record the operation log of each task unit, including input and output data and execution time.

[0184] In the embodiment of the present invention, the dynamic operation of the execution workflow ensures the step-by-step execution of the task unit and the real-time generation of the intermediate result data. The accuracy of the data transmission path and the comprehensive record of the operation log provide support for the transparency and traceability of the execution of the task unit. The recorded information of the execution time and input and output data not only provides a basis for result verification, but also accumulates data support for the performance optimization of the task chain.

[0185] In a preferred embodiment of the present invention, detecting whether the intermediate result data meets the requirements of the logical relationship and content style, and adjusting the inconsistent parts to generate task unit optimization data includes:

[0186] Verify the format, logic and content style of each intermediate result data, and obtain the data that failed the verification and the task units to be adjusted;

[0187] According to the data that failed the verification, the execution rules of the task unit to be adjusted or the tool adaptation parameter problem are analyzed to obtain the analysis result data;

[0188] According to the analysis result data, the execution rule parameters of the task unit to be adjusted are adjusted or the tools are reallocated, and the task unit to be adjusted is re-run to generate task unit optimization data.

[0189] In the embodiment of the present invention, the logical relationship and content style detection of the intermediate result data provide a key link for the quality control of the method. By analyzing and adjusting the data that failed the verification, the optimized task unit data can further improve the quality and consistency of the output results. This adjustment mechanism ensures the flexibility of the method in handling complex task chains, while enhancing the fault tolerance and reliability of the method in creative scene applications.

[0190] More specifically, the format, logic, and content style of each intermediate result data are verified to obtain the data that failed the verification and the task units to be adjusted. The specific steps include:

[0191] The goal of format verification is to ensure that the intermediate result data meets the technical specifications and can be used normally by subsequent processing programs. Check whether the text encoding format such as UTF-8, paragraph structure, title hierarchy, and symbol usage meet the regulations. For example, if the user requires the output copy to have a hierarchical title format, it is necessary to verify whether the title, subtitle, and paragraph structure are included. Check whether the image resolution, color mode such as RGB or CMYK, and file format such as PNG and JPEG meet the requirements. For example, if the image is targeted for printing, it must comply with the CMYK color mode. Check whether the video and audio encoding formats such as MP4, AAC, resolution, and frame rate meet the requirements. For example, verify whether the video is 1920x1080 resolution and whether the audio meets the specified sampling rate.

[0192] The goal of logical verification is to ensure that the semantic associations between the data contents of the intermediate results are correct and that the data dependencies are consistent. Verify whether the output meets the logical requirements of the input. For example, if the output of task unit A is the input of task unit B, it is necessary to check whether task unit B correctly uses the data of A. Check whether the generated data is consistent with the context. For example, if a certain text describes the target user group as "young people", then the related images and video content must have a style or elements that are consistent with the characteristics of "young people". Check whether the output of the task unit is consistent with other units. For example, if the copy and design drawings are used together for marketing activities, it is necessary to verify whether the product features described in the copy are consistent with the design. Figure 1 To.

[0193] The goal of style verification is to ensure that the generated content meets the user's style requirements, including language style, visual style, etc. Check whether the copy meets the target tone, such as formal, humorous, concise, etc., and ensure that the wording and sentence structure are consistent with the user's brand style. For example, the tone of marketing copy may need to be persuasive, while the copy of academic reports needs to be neutral and rigorous. Check whether the image meets the expected artistic style or brand design specifications. For example, corporate brands may require a flat design, while advertising images may require bright and eye-catching colors. Check whether the color tone, editing rhythm, and background music of the video are consistent with the target style. For example, videos for casual brands may require a relaxed and bright style, while videos for high-end brands may require a sophisticated and calm performance.

[0194] Data that failed verification refers to data that does not meet the requirements during the above verification process. For example, the text contains an incorrect encoding format or the paragraph does not meet expectations. The image resolution is too low or the color mode does not meet the requirements. The logical content of the video is inconsistent, such as the description of the timeline conflicts with the picture.

[0195] By verifying the data that failed, you can determine the specific task unit that needs to be adjusted. The location process includes: Checking which task unit generated the failed data. For example, if the image format verification fails, locate the task unit that generated the image. Confirm whether the rules or input parameters of the task unit lead to incorrect output. For example, if the multimedia format does not meet expectations, you need to check whether the encoding settings of the task unit are correct.

[0196] The verification process can be achieved through the following automated tools: For example, text format verification can use regular expressions to detect paragraph structure; image format verification can use image processing libraries such as OpenCV to check resolution and color mode. Natural language processing tools such as BERT can be used for semantic consistency analysis; data flow tracking tools can check whether the input and output of task units match. Text style verification can be completed through style transfer models such as GPT; image and multimedia style verification can use deep learning models such as StyleGAN to detect whether the content meets the style requirements.

[0197] By verifying the format, logic, and content style of each intermediate result data, this method can discover potential problems during the execution of the task chain and locate the task units to be adjusted in a timely manner. The processing of failed data and the optimization of task units are verified to ensure the high quality and consistency of the final generated results. This process avoids the complexity of traditional manual adjustments and improves the level of automation and output efficiency. Ultimately, this verification mechanism can significantly improve the reliability of generated content and user satisfaction.

[0198] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. The AI ​​workflow automation management method based on creative scenarios is characterized by: The method comprises: Acquiring initial data input by a user, wherein the initial data includes text data describing a target creative scene and non-text reference data; Based on the initial data, key elements are extracted from the text data, and relevant features are analyzed from the non-text reference data to obtain parsed data; Constructing an initial task chain according to the parsed data, wherein the initial task chain includes a plurality of task units and their logical association paths; According to the logical association path, the dependency relationship of the task unit is analyzed to generate a set of dependency paths. The dependency path is used to describe the direct data transfer and execution order between the task units. The dependency relationships between task units in the dependency path set are weighted and priorities are assigned to the results to generate a dynamic path table, which includes the execution order of task units, the transmission order of dependency data, and adjustable backup paths. Optimize the initial task chain according to the dynamic path table and generate a dynamic task chain table; According to the execution requirements of the task units in the dynamic task list, select the appropriate tools from the AI ​​tool library, allocate corresponding tool resources to each task unit, and generate matching data; Constructing an execution workflow according to the matching data, wherein the execution workflow includes the execution order and data transfer path of the task units, and associating the execution rules of each task unit with the corresponding tool; Dynamically run the execution workflow to obtain intermediate result data, wherein the intermediate result data includes the content generated by each task unit; Check whether the intermediate result data meets the requirements of logical relationship and content style, adjust the inconsistent parts, and generate task unit optimization data; Classify, format and check the consistency of task unit optimization data according to the target requirements of the execution workflow, and generate output result data, which includes text, images or multimedia content generated according to the user's target requirements; The initial task chain is optimized according to the dynamic path table to generate a dynamic task chain table, including: Extracting task unit information from the dynamic path table, including the execution order of the task units, the transmission path of dependent data, and the backup path information; According to the transmission path of dependent data, the logical association path in the initial task chain is adjusted to form an optimized task path; According to the backup path information, a redundant execution path is allocated to the task units in the initial task chain; Integrate the optimized task path and redundant execution path, rearrange the execution order of task units in the initial task chain, and generate a dynamic task chain table; the redundant execution path is used as a backup path as an alternative solution when the main path is abnormal; According to the content of the dynamic task chain table, the execution rules and path information of each task unit are updated to the task dependency relationship, forming a dependency record including the execution order of the task units, the dependency weight value and the path redundancy scheme; According to the execution requirements of the task units in the dynamic task list, select the appropriate tools from the AI ​​tool library, allocate corresponding tool resources to each task unit, and generate matching data, including: Extract the execution requirements of task units from the dynamic task list, including input data format, output target and processing time limit; According to the execution requirements of the task unit, the tool resource with the highest adaptability is allocated from the AI ​​tool library, and the tool allocation information is generated. The tool allocation information includes the task unit identifier, tool identifier, and tool adaptation parameters. The calculation formula of the adaptability is as follows: ， in, Task Unit The adaptability, For tools The adaptability of the tool Satisfy the task unit The ability to execute requirements, For tools The average processing time, Task Unit The upper limit of the processing time indicates the time constraint of the task. Task Unit The execution priority reflects the priority of the task unit in the dynamic task list. Task Unit The resource demand degree indicates the number of tool resources that the task unit needs to call. Task Unit The tool switching cost represents the time or resource cost of switching from other tools to the current tool. For tools The weight coefficient of and is the adjustment coefficient, is the number of tools; Input the tool allocation information into the dynamic task chain table to generate the tool calling path of the task unit; Matching data is formed according to the tool allocation information and the tool calling path.

2. The AI ​​workflow automation management method based on creative scenarios according to claim 1 is characterized in that: Based on the initial data, key elements are extracted from the description information and relevant features are analyzed from the reference content to obtain parsed data, including: According to the text data, the text sentences are decomposed, the keywords are located, and the first content data is extracted by analyzing the context relevance; By content segmentation, local correlation characteristics of non-text reference data are marked and classified, feature extraction is performed, and second content data is generated; grouping the first content data and the second content data according to category characteristics to generate characteristic grouping data; Group data according to each feature, extract correlation features based on the cross-matching method, and build a multi-layer data association network; Filter subset data that meet the preset correlation threshold from the multi-layer data association network to generate parsed data.

3. The AI ​​workflow automation management method based on creative scenarios according to claim 2 is characterized in that: Based on the parsed data, build the initial task chain, including: Extracting information elements of the task unit from the parsed data, the information elements including task objectives, input data formats, and expected output types; Analyze information elements and build logical association paths based on the dependencies between task objectives; Use logical association paths to mark the data flow and execution order between task units and generate task dependencies; According to the task dependency, the task units are arranged in sequence according to the logical association path to build the initial task chain; Assign a unique identifier to each task unit in the initial task chain to record its task objectives, input data sources, and output target destinations.

4. The AI ​​workflow automation management method based on creative scenarios according to claim 3 is characterized in that: According to the logical association path, the dependency relationship of the task unit is analyzed and a dependency path set is generated, including: Extract the logical association path of each task unit from the initial task chain, and establish a direct dependency path based on the data input and output relationship between task units; The upstream and downstream task units of each direct dependency path are classified to form a set of strong dependency paths and weak dependency paths. If the downstream task unit is highly dependent on the output of the upstream task unit and cannot be replaced by other paths, the path is considered a strong dependency path. If the downstream task unit's dependence on the upstream task unit can be replaced by other paths or redundant methods, the path is considered a weak dependency path. Analyze the interaction frequency of task units in the strong dependency path set and calculate the dependency strength between task units; According to the dependency strength, priorities are assigned to the task units in the strong dependency path set, and a dependency path set is generated. The dependency path set includes direct data transfer and execution order between task units.

5. The AI ​​workflow automation management method based on creative scenarios according to claim 4 is characterized in that: The dependency relationships between task units in the dependency path set are weighted and the results are assigned priorities to generate a dynamic path table, including: Obtain the task unit information of each path from the dependent path set, including the input data source, output target destination and dependency strength of the task unit; According to the data flow frequency, path length and data processing format of the task unit, the dependency relationship between the task units in the dependency path set is weighted to obtain the dependency weight value; the calculation formula of the dependency weight value is as follows: ， in, Task Unit With task unit The dependency weight between Task Unit With task unit The frequency of data flow between Task Unit With task unit The path length between Task Unit With task unit The format matching between Task Unit With task unit The path abnormality rate between , indicates the probability of interruption or delay of the path, Task Unit With task unit The conflict rate between them indicates the frequency of task resource conflicts. , , , , and is the weight coefficient; The dependency weight values ​​are divided according to the preset dependency thresholds to generate a dependency path priority table, which is used to optimize the execution order of the task units; According to the dependency path priority table, the execution order of the task units is arranged to generate a dynamic path table, which includes the execution order of the task units, the transmission path of the dependency data and the backup path information.

6. The AI ​​workflow automation management method based on creative scenarios according to claim 5 is characterized in that: Build and execute workflows based on matching data, including: Extract the execution order and tool calling path of each task unit from the dynamic task linked list; According to the execution order of task units, establish the data transmission path between task units; The execution workflow is constructed based on the tool calling path and data transmission path. The execution workflow includes the execution rules of the task unit, data flow and tool calling relationship.

7. The AI ​​workflow automation management method based on creative scenarios according to claim 6 is characterized in that: Dynamically run the execution workflow and obtain intermediate result data, including: Run the task units in the execution order of the task units in the execution workflow, and call the corresponding tool resources to process the input data during the running process; After each task unit completes execution, it generates intermediate result data and transmits the result to the subsequent task unit through the data transmission path; Record the operation log of each task unit, including input and output data and execution time.

8. The AI ​​workflow automation management method based on creative scenarios according to claim 7 is characterized in that: Check whether the intermediate result data meets the requirements of logical relationship and content style, adjust the inconsistent parts, and generate task unit optimization data, including: Verify the format, logic and content style of each intermediate result data, and obtain the data that failed the verification and the task units to be adjusted; According to the data that failed the verification, the execution rules of the task unit to be adjusted or the tool adaptation parameter problems are analyzed to obtain the analysis result data; According to the analysis result data, the execution rule parameters of the task unit to be adjusted are adjusted or the tools are reallocated, and the task unit to be adjusted is re-run to generate task unit optimization data.

Images