Workflow optimization method and apparatus based on cognitive feedback and semantic tracing

Abstract

The application provides a workflow optimization method and device based on cognitive feedback and semantic traceability, and relates to the field of workflow processing, and solves the technical problem of low content generation efficiency of the existing workflow mode. The method comprises the following steps: in response to an input operation for a user demand, generating initial demand feedback content according to a target user demand corresponding to the input operation; detecting the initial demand feedback content based on a preset cognitive detection model to generate cognitive defect data of the initial demand feedback content; reversely deducing the generation of the cognitive defect data through a traceability reasoning engine to obtain a semantic traceability result of the cognitive defect data, and mapping the cognitive defect data back to a workflow in a generation process of the initial demand feedback content; dynamically adjusting a path structure of the workflow according to the semantic traceability result, iteratively reconstructing demand feedback content corresponding to the target user demand based on the path structure of the adjusted workflow, and obtaining final demand feedback content.

Description

Technical Field

[0001] This application relates to the field of workflow processing technology, and in particular to a workflow optimization method and apparatus based on cognitive feedback and semantic abduction. Background Technology

[0002] Currently, existing AI generation workflow technologies mostly focus on optimizing the model itself or simple chained task pipelines. A common approach is to use user input prompts, with a large generative model (such as LLM or diffusion models) generating the output. If the generated content is unsatisfactory, the user needs to manually adjust the prompts and resubmit. Another approach is to pre-set a task sequence, such as first generating text, then generating images based on the text, and finally synthesizing audio and video, but the process is fixed. Therefore, this existing workflow method has low content generation efficiency. Summary of the Invention

[0003] The purpose of this invention is to provide a workflow optimization method and apparatus based on cognitive feedback and semantic abduction to solve the technical problem of low content generation efficiency in existing workflow methods.

[0004] Firstly, this application provides a workflow optimization method based on cognitive feedback and semantic abduction, the method comprising: In response to an input operation that addresses a user's needs, an initial feedback content is generated based on the target user's needs corresponding to the input operation. Based on the initial demand feedback content, a preset cognitive detection model is used to detect and generate cognitive defect data of the initial demand feedback content; The cognitive defect data is reverse-engineered through the abductive reasoning engine to obtain the semantic abductive result of the cognitive defect data, and the cognitive defect data is mapped back to the workflow of the initial demand feedback content in the generation process. The workflow path structure is dynamically adjusted based on the semantic causation results. The requirement feedback content corresponding to the target user requirement is iteratively reconstructed based on the adjusted workflow path structure to obtain the final requirement feedback content.

[0005] In one possible implementation, the step of detecting cognitive deficiency data of the initial demand feedback content using a preset cognitive detection model includes: Based on the user intent corresponding to the target user's needs, retrieve all relevant data and determine whether the target user's needs are original based on the quantity of the relevant data and its relevance to the user intent. Since the target user's need is not an original need, the main learning time period of the preset cognitive detection model is determined based on the generation time of the relevant data; different learning time periods correspond to different cognitive content; Based on the target cognitive content corresponding to the main learning time period, the preset cognitive detection model is cognitively optimized to obtain the first target cognitive detection model. Based on the initial demand feedback content, the first target cognitive detection model is used to detect and generate cognitive defect data of the initial demand feedback content.

[0006] In one possible implementation, after determining whether the target user's need is an original need based on the quantity of the relevant data and its relevance to the user's intent, the method further includes: Since the target user's need is an original need, the target user type is identified based on the natural language description content corresponding to the input operation and the user intent corresponding to the target user's need; the user type includes at least one of user age, user industry field, user occupation category, and user professional direction; different user types correspond to different cognitive directions and business logics. Based on the target business logic and target cognitive direction corresponding to the target user type, the preset cognitive detection model is optimized in a logical and cognitive manner to obtain a second target cognitive detection model. Based on the initial demand feedback content, the second target cognitive detection model is used to detect and generate cognitive defect data of the initial demand feedback content.

[0007] In one possible implementation, the step of generating initial request feedback content based on the target user request corresponding to the input operation in response to the user request includes: In response to input operations that meet user needs, a large-scale integrated language model is used to perform deep semantic understanding and user need task parsing on the natural language description content corresponding to the input operation, so as to obtain the deep intent, core theme, sentiment tendency, style requirements, implicit constraints and key entities of the user need task. Based on the underlying intent, the core theme, the emotional inclination, the style requirements, the implicit constraints, and the key entities, a structured initial requirement feedback blueprint data is generated. The initial requirement feedback blueprint data is transformed into refined instructions through a multimodal generation module, and the instructions drive the corresponding text model, image model and / or audio model to generate the initial requirement feedback content.

[0008] In one possible implementation, the step of detecting cognitive deficiency data of the initial demand feedback content using a preset cognitive detection model includes: The accuracy and consistency of factual statements in the initial demand feedback content are verified using a search engine based on the knowledge graph corresponding to the preset cognitive detection model. In response to the factual statements meeting the preset accuracy and consistency requirements, a preset cognitive detection model is used to analyze the logical coherence of the argumentation structure and causal chain of the initial demand feedback content to determine whether there are any logical fallacies. In response to the absence of logical fallacies, determine whether the emotional tone and style of the initial requirement feedback content are consistent with the initial requirement feedback blueprint data; In response to the initial requirement feedback blueprint data that meets the requirements, the semantic matching degree of the initial requirement feedback content is checked through cross-modal consistency verification to see if it meets the preset matching degree. In response to the semantic matching degree not meeting the preset matching degree, cognitive defect data of the initial demand feedback content in the semantic matching dimension is generated.

[0009] In one possible implementation, the cognitive deficit data refers to the factual statements in the initial demand feedback content not meeting the preset accuracy requirement; The step of reverse-engineering the generation of the cognitive deficit data through abductive reasoning engine to obtain the semantic abductive result of the cognitive deficit data includes: By using abductive reasoning engine to reverse-engineer the generation of the cognitive defect data, it is determined that the root cause of the semantic abduction result of the cognitive defect data is that the contextual definition of key entities in the initial demand feedback content is ambiguous, which leads to the generation model activating incorrect knowledge associations.

[0010] In one possible implementation, the dynamic adjustment of the workflow path structure based on the semantic ablation result, and the iterative reconstruction of the requirement feedback content corresponding to the target user requirement based on the adjusted workflow path structure, to obtain the final requirement feedback content, includes: Based on the semantic abduction results, structured decisions are generated through the workflow scheduling center, and the semantic abduction results are transformed into optimization instructions; In response to the detection that the semantic causation result contains a knowledge blind spot, the knowledge retrieval and integration sub-process node corresponding to the structured decision is inserted into the workflow based on the optimization instruction to obtain the path structure of the adjusted workflow; In response to the detection that the semantic causation result contains an error in the deconstruction of the user's demand task, the structured decision is fed back to the semantic understanding sub-process node in the workflow based on the optimization instruction, so as to reconstruct and optimize the deconstruction model of the user intent in the semantic understanding sub-process node. The above steps are executed repeatedly, and in each iteration, the workflow path structure is iterated based on the cognitive deficit data and the semantic causal results, until there is no cognitive deficit data in all dimensions of the requirement feedback content, and the final requirement feedback content is output.

[0011] Secondly, this application provides a workflow optimization device based on cognitive feedback and semantic abduction, comprising: The generation module is used to respond to input operations that address user needs and generate initial feedback content based on the target user needs corresponding to the input operations. The detection module is used to detect the initial demand feedback content using a preset cognitive detection model and generate cognitive defect data of the initial demand feedback content. The causal module is used to reverse deduce the generation of the cognitive defect data through the causal reasoning engine, obtain the semantic causal result of the cognitive defect data, and map the cognitive defect data back to the workflow path of the initial demand feedback content in the generation process. The reconstruction module is used to dynamically adjust the workflow path based on the semantic causation results, and iteratively reconstruct the requirement feedback content corresponding to the target user requirement based on the adjusted workflow path to obtain the final requirement feedback content.

[0012] Thirdly, this application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the method described in the first aspect above.

[0013] Fourthly, this application also provides a computer-readable storage medium storing computer-executable instructions that, when invoked and executed by a processor, cause the processor to perform the method described in the first aspect above.

[0014] This application brings the following beneficial effects: This application provides a workflow optimization method and apparatus based on cognitive feedback and semantic abduction. It is capable of responding to input operations targeting user needs, generating initial demand feedback content based on the target user needs corresponding to the input operation; detecting cognitive deficiencies in the initial demand feedback content using a preset cognitive detection model to generate cognitive defect data; reverse-engineering the generation of the cognitive defect data using a abduction reasoning engine to obtain semantic abduction results for the cognitive defect data, and mapping the cognitive defect data back to the workflow during the generation of the initial demand feedback content; dynamically adjusting the path structure of the workflow based on the semantic abduction results, and iteratively reconstructing the demand feedback content corresponding to the target user needs based on the adjusted workflow path structure. The final feedback content is obtained. In this solution, through cognitive detection, abductive reasoning, and dynamic adjustment of the workflow path structure, an intelligent generation workflow system with autonomous cognitive feedback and abductive optimization capabilities is realized. This enables the workflow not only to generate content but also to detect defects in the output content, deeply deduce the root causes of defects, and autonomously formulate and execute optimization strategies based on these defects, i.e., formulate precise strategies to improve the workflow. This system breaks the traditional linear pipeline workflow generation mode and establishes a dynamic intelligent system with a cognitive closed loop as its core, thereby realizing a paradigm shift in the system's autonomous iterative optimization. It achieves refined and intelligent management of the entire generation workflow process, significantly improves the quality and efficiency of content generation, and solves the technical problem of low content generation efficiency in existing workflow methods.

[0015] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating the workflow optimization method based on cognitive feedback and semantic abduction provided in this application embodiment; Figure 2 Another flowchart illustrating the workflow optimization method based on cognitive feedback and semantic abduction provided in this application embodiment; Figure 3 A schematic diagram of a workflow optimization device based on cognitive feedback and semantic abduction provided in this application embodiment; Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] The terms "comprising" and "having," and any variations thereof, used in the embodiments of this application, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the steps or units listed, but may optionally include other steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0020] Currently, the common approach in existing technologies involves user input prompts, with large generative models generating the output. If the result is unsatisfactory, the user must manually adjust the prompts and resubmit, forming an open loop of "generation-evaluation-manual modification." Another approach is to pre-set a task sequence, such as first generating text, then generating images based on the text, and finally synthesizing audio and video. However, this process is fixed and lacks the ability to perceive and dynamically adjust intermediate results. These methods are essentially passive trial-and-error mechanisms that rely heavily on human intervention.

[0021] These existing methods are extremely inefficient, requiring users to possess specialized engineering knowledge and engage in repeated manual debugging, consuming significant time and computing resources. Secondly, the workflow is rigid, unable to intelligently route and make decisions based on the real-time quality of the generated content, causing errors to accumulate and amplify along the pipeline. For example, a piece of off-topic copy can render subsequent generated images completely useless. More importantly, existing technologies lack deep semantic understanding and causal reasoning capabilities. The system cannot understand "why the output is not ideal," and therefore cannot autonomously and accurately locate the root cause of the problem and implement effective corrective strategies, resorting only to blindly repeating generation or relying entirely on human guidance.

[0022] Based on this, the present application provides a workflow optimization method and apparatus based on cognitive feedback and semantic attribution, which can solve the technical problem of low content generation efficiency in existing workflow methods.

[0023] The embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0024] Figure 1 This is a flowchart illustrating a workflow optimization method based on cognitive feedback and semantic abduction, provided as an embodiment of this application. Figure 1 As shown, the method includes: Step S110: In response to an input operation targeting user needs, generate initial requirement feedback content based on the target user needs corresponding to the input operation.

[0025] As one possible implementation, the above-mentioned response to an input operation targeting a user's needs, generating initial feedback content based on the target user's needs corresponding to the input operation, may specifically include the following steps: In response to user input operations, a large integrated language model is used to perform deep semantic understanding and user demand task parsing on the natural language description content corresponding to the input operation, so as to obtain the deep intent, core theme, sentiment tendency, style requirements, implicit constraints and key entities of the user demand task. Structured initial requirement feedback blueprint data is generated based on deep intent, core themes, sentiment, style requirements, implicit constraints, and key entities. The initial requirement feedback blueprint data is transformed into refined instructions through a multimodal generation module, and the instructions drive the corresponding text model, image model, and / or audio model to generate initial requirement feedback content.

[0026] For example, such as Figure 2 As shown, firstly, the semantic understanding and task deconstruction module performs deep semantic analysis on the user's input natural language description. This goes beyond simply extracting keywords; instead, it uses an integrated large-scale language model to extract the task's underlying intent, core themes, sentiment, stylistic requirements, implicit constraints, and key entities (such as people, places, and concepts). Ultimately, it outputs a structured "generation blueprint," providing precise and computable standards and roadmaps for subsequent generation and evaluation. For example, if the user inputs "write an optimistic, technology-oriented blog post about sustainable energy," the module will output structured semantic elements, providing a benchmark for all subsequent stages.

[0027] Next, the multimodal generation module receives this "generation blueprint" and transforms it into highly refined instructions, driving the corresponding text, image, or audio models to generate initial content. The input quality in this step is far superior to the user's original input because it is already a structured, unambiguous semantic representation.

[0028] Step S120: Based on the initial demand feedback content, a preset cognitive detection model is used to detect and generate cognitive defect data of the initial demand feedback content.

[0029] In one optional implementation, the above-mentioned detection based on the initial demand feedback content using a preset cognitive detection model to generate cognitive defect data of the initial demand feedback content may specifically include the following steps: Based on the user intent corresponding to the target user's needs, retrieve all relevant data and determine whether the target user's needs are original based on the quantity of relevant data and the degree of relevance to the user intent. In response to the non-original needs of the target users, the main learning time period of the preset cognitive detection model is determined based on the generation time of relevant data; different learning time periods correspond to different cognitive content; Based on the target cognitive content corresponding to the main learning time period, the preset cognitive detection model is optimized for cognitive specificity to obtain the first target cognitive detection model; based on the initial demand feedback content, the first target cognitive detection model is used for detection to generate cognitive defect data of the initial demand feedback content.

[0030] For example, the system first receives user input regarding a specific need through a human-computer interaction interface, parses the input, extracts the core target user need, and then generates an initial feedback document as the baseline text for subsequent analysis. Next, the system activates the intent recognition module to deeply analyze the user intent behind the target user need, and uses this as a guide to retrieve all relevant data from internal databases and external networks. The system then quantitatively evaluates the retrieved data, comprehensively considering the quantity and semantic relevance to the user intent, to determine whether the target user need is a non-original need with existing precedents or a completely new and original need.

[0031] When the system determines that the target user's need is not original, it will enter the model optimization phase. The system will further analyze the generation time distribution of the retrieved relevant data to determine the main learning period for the preset cognitive detection model, as industry cognitive standards and technical levels differ across time periods. The system then retrieves the target cognitive content corresponding to this main learning period (such as industry standards, technical documents, or historical cases at the time), and uses this time-sensitive knowledge to fine-tune or incrementally learn the preset cognitive detection model, completing targeted cognitive optimization and constructing a first target cognitive detection model adapted to the current need context.

[0032] Finally, the system enters the detection and execution phase. The initially generated feedback content is input into the newly optimized first-target cognitive detection model. This model performs a deep scan and logical comparison of the feedback content based on cognitive standards over a specific time period, identifying knowledge blind spots, logical loopholes, or content that does not conform to the current cognitive level. Ultimately, it outputs detailed cognitive defect data, providing accurate data support for subsequent requirement corrections or feedback optimizations.

[0033] By employing a closed-loop process of dynamic retrieval, time-based positioning, and targeted model optimization, the problem of cognitive lag and standard misalignment in general cognitive models when dealing with specific domains or time periods is solved, thus eliminating the cognitive time lag of general models.

[0034] Specifically, traditional cognitive detection models are usually trained on large-scale data over a fixed period of time, and their cognition often remains in the past. When faced with non-original needs (i.e., existing historical data or specific industry background), general models may not be able to understand the specific context or standards of the need in a specific time period. However, the solution in this application determines the main learning period by retrieving the generation time of relevant materials, and optimizes the model cognitively accordingly. This is equivalent to giving the model temporary tutoring, enabling it to have a sense of the times and professionalism that matches the target needs.

[0035] Furthermore, different user needs may correspond to different industry standards or historical backgrounds. Using the same standard to test all needs can easily lead to misjudgments (for example, misjudging correct practices under old standards as defects). In this application, instead of using a single fixed model, a first target cognitive detection model is dynamically generated based on the target user's needs. This means that the detection standard is dynamically adjusted according to user intent and relevant data, thereby ensuring that the generated cognitive defect data is more accurate, objective, and consistent with the actual logic under that specific need, achieving a customized detection standard for each individual.

[0036] In this embodiment, by retrieving relevant information based on the target user's needs and determining the main learning time period, the preset cognitive detection model is optimized in a targeted manner, achieving dynamic updating and adaptation of the detection model's cognitive capabilities. This enables the system to detect initial demand feedback based on cognitive content that is highly relevant to the user's intent and has timeliness, effectively avoiding detection bias caused by outdated or overly generalized model training data, and significantly improving the accuracy and relevance of cognitive defect data identification.

[0037] In another alternative implementation, after determining whether the target user's need is an original need based on the quantity of relevant data and its relevance to the user's intent, the method may further include the following steps: Responding to the original needs of the target users, the target user type is identified based on the natural language description of the input operation and the user intent corresponding to the target user's needs. The user type includes at least one of the following: user age, user industry field, user occupation category, and user professional direction. Different user types correspond to different cognitive directions and business logic. Based on the target business logic and target cognitive direction corresponding to the target user type, the preset cognitive detection model is optimized in terms of logic and cognition to obtain the second target cognitive detection model; based on the initial demand feedback content, the second target cognitive detection model is used to detect and generate cognitive defect data of the initial demand feedback content.

[0038] For example, upon recognizing a target user's request as original, the system first initiates a deep analysis of user attributes. The system performs natural language processing on the input that triggered the request, extracting semantic features from the text and combining this with the user's intent behind the request to perform multi-dimensional user profiling, thereby determining the target user type. This identification process specifically encompasses the precise capture of at least one of the following: user age group, industry domain characteristics, occupational category attributes, and professional research direction, aiming to clarify who made the request.

[0039] After identifying the target user type, the system invokes the target business logic and target cognitive direction pre-associated with that type. Because users from different backgrounds (e.g., senior experts versus beginners, or financial professionals versus artists) exhibit significant differences in their thinking patterns and business processes, the system utilizes these specific logic and cognitive parameters to perform in-depth logical construction and cognitive alignment optimization of the preset cognitive detection model. This process aims to adjust the model's evaluation criteria, switching it from a general mode to a dedicated mode adapted to this specific user type, thereby constructing a second target cognitive detection model.

[0040] Finally, the system performs targeted detection tasks. The previously generated initial feedback is fed into a customized and optimized second-target cognitive detection model. This model performs a deep scan of the feedback based on specific business logic and cognitive perspectives adapted to the target user type, identifying logical gaps, misuse of terminology, or defects that do not conform to the cognitive habits of this type of user. Ultimately, it outputs cognitive defect data that accurately reflects cognitive biases from the specific user's perspective.

[0041] In this embodiment, the cognitive detection model is personalized and context-aligned for different user groups, significantly improving the semantic understanding accuracy and logical relevance of feedback on original needs. Specifically, for original needs lacking historical data references, general detection models often struggle to grasp their specific business context, easily leading to problems such as "misjudgment due to lack of expertise" or "mismatch of logical standards." This solution dynamically adjusts the detection model's "thinking mode" (i.e., business logic and cognitive direction) by identifying the user's specific type (e.g., industry, occupation, age), enabling it to examine the feedback from the perspective of that specific user group. This mechanism ensures that the generated cognitive defect data is not merely grammatical or general logical correction, but also in-depth business logic verification that aligns with the user's specific professional background and cognitive habits, thereby greatly enhancing the effectiveness of the feedback and user acceptance.

[0042] As an optional implementation, the above-mentioned detection based on the initial demand feedback content using a preset cognitive detection model to generate cognitive defect data of the initial demand feedback content may specifically include the following steps: The accuracy and consistency of factual statements in the initial feedback content are verified using a search engine based on the knowledge graph corresponding to the pre-set cognitive detection model. In response to the factual statements meeting the pre-set accuracy and consistency requirements, the logical coherence of the argumentation structure and causal chain of the initial feedback content is analyzed using the pre-set cognitive detection model to determine whether there are any logical fallacies. In response to the absence of logical fallacies, determine whether the emotional tone and style of the initial requirement feedback content conform to the initial requirement feedback blueprint data; in response to conformity with the initial requirement feedback blueprint data, check whether the semantic matching degree of the initial requirement feedback content conforms to the preset matching degree through cross-modal consistency verification; in response to the semantic matching degree not conforming to the preset matching degree, generate cognitive defect data of the initial requirement feedback content in the semantic matching dimension.

[0043] For example, such as Figure 2 As shown, the generated content is then sent to the cognitive detection module ( Figure 2 The cognitive assessment module (in the context of the content) goes far beyond traditional content moderation or simple scoring. It incorporates multiple advanced models for in-depth content quality assessment: (1) Fact consistency check: Use knowledge graphs or search engines to verify the accuracy of factual statements in the generated content; (2) Logical coherence analysis: Analyze whether the argument structure and causal chain are reasonable and whether there are logical fallacies; (3) Emotional and stylistic consistency: Determine whether the emotional tone and style of the content are consistent with the requirements in the "Generation Blueprint"; (4) Aesthetic and technical requirements (for images / videos): Evaluate professional indicators such as composition, color, and lighting; (5) Cross-modal consistency verification: Ensure that the content of images, texts, audio and video is highly matched semantically.

[0044] The output of the cognitive assessment module is not a simple score, but a detailed assessment report that clearly points out the specific dimensions in which the content is deficient.

[0045] Step S130: The generation of cognitive defect data is reversed through the abductive reasoning engine to obtain the semantic abductive result of cognitive defect data, and the cognitive defect data is mapped back to the workflow of the initial demand feedback content in the generation process.

[0046] For example, the aforementioned cognitive deficit data refers to the factual statements in the initial feedback content not meeting the preset accuracy requirements. The aforementioned reverse derivation of the generation of cognitive deficit data using abductive reasoning engines to obtain semantic abductive results for the cognitive deficit data may specifically include the following steps: By using abductive reasoning engine to reverse-engineer the generation of cognitive defect data, the root cause of the semantic abduction results of cognitive defect data is determined to be that the contextual definition of key entities in the initial demand feedback content is ambiguous, which leads to the generative model activating incorrect knowledge associations.

[0047] For example, such as Figure 2 As shown, the aforementioned inspection and evaluation report is sent to the intelligent core of the entire system, namely the abductive reasoning engine. The engine's task is to answer a key question: "Why do these defects occur?" Through reverse reasoning, it maps quality defects back to a specific stage in the generation process. For example, if a blog post contains factual errors, the abductive engine might determine that "the ambiguous contextual definition of the key entity 'fusion energy' in the 'generation blueprint' caused the generation model to activate incorrect knowledge associations." This ability to trace the source is fundamental to this solution's superiority over simple retry mechanisms.

[0048] Through deep semantic evaluation and abductive reasoning, the system transcends superficial, statistically-based quality scoring. It can diagnose deep semantic flaws in generated content (such as logical fallacies, factual errors, and emotional conflicts) and deduce the root causes of these flaws. This is equivalent to giving the system the ability to "not only know what," but also "why," enabling it to accurately pinpoint the crux of the problem rather than blindly trying different approaches.

[0049] Step S140: Dynamically adjust the workflow path structure based on the semantic causation results, and iteratively reconstruct the requirement feedback content corresponding to the target user's requirements based on the adjusted workflow path structure to obtain the final requirement feedback content.

[0050] like Figure 2 As shown, in the core "generation-evaluation-abortion-decision" closed loop of the system, the workflow scheduling center is the heart, responsible for pumping instructions and data; the abductive reasoning engine is the brain, responsible for thinking; the cognitive evaluation module is the sensory system, responsible for checking; and the multimodal generation module is the executive organ.

[0051] For example, the above-mentioned dynamic adjustment of the workflow path structure based on semantic ablation results, and the iterative reconstruction of the requirement feedback content corresponding to the target user's requirements based on the adjusted workflow path structure to obtain the final requirement feedback content, may specifically include the following steps: Based on the semantic abduction results, structured decisions are generated through the workflow scheduling center, and the semantic abduction results are transformed into optimization instructions. In response to the detection that there are knowledge blind spots in the semantic abduction results, the knowledge retrieval and integration sub-process nodes corresponding to the structured decisions are inserted into the workflow based on the optimization instructions to obtain the adjusted workflow path structure. In response to the detection of errors in the deconstruction of user demand tasks in the semantic causation results, the structured decision is fed back to the semantic understanding sub-process nodes in the workflow based on optimization instructions, so as to reconstruct and optimize the deconstruction model of user intent in the semantic understanding sub-process nodes. The above steps are executed repeatedly, and in each iteration, the workflow path structure is iterated based on cognitive deficit data and semantic attribution results until there is no cognitive deficit data in all dimensions of the requirement feedback content, and the final requirement feedback content is output.

[0052] Dynamic workflow reconstruction based on abductive reasoning results. The system's decisions are no longer a simple binary choice of "retry" or "continue," but rather a dynamic adjustment of the workflow's structure based on the semantic root causes obtained through abduction. For example, if abduction reveals that the generation failure is due to insufficient knowledge, the system will automatically insert a "knowledge retrieval" node; if it's due to instruction ambiguity, it will initiate a "semantic clarification" sub-dialogue. The workflow is no longer a pre-defined pipeline, but an organic intelligent entity that can be structurally reshaped based on real-time cognitive feedback.

[0053] like Figure 2 As described above, in the deep interaction and workflow of the core modules, the operation of the entire system begins with a relatively simple description of the user's intent. After receiving the task, the system does not mechanically execute it in a fixed order, but instead initiates an intelligent generation-evaluation-decision loop.

[0054] For example, based on abductive conclusions, the workflow scheduling hub (dynamic rule engine) makes precise, structured decisions. These decisions are no longer simply about "retrying," but rather include: (1) Instruction optimization and regeneration: Transform the causal conclusions into precise optimization instructions and feed them back to the generation module.

[0055] (2) Dynamic adjustment of workflow path: Decision inserts new nodes, such as when the cause is traced to find a "knowledge blind spot", the "knowledge retrieval and integration" sub-process is automatically inserted.

[0056] (3) Generate blueprint iterative reconstruction: If the cause is found to be an error in the initial task deconstruction, it will be fed back to the semantic understanding module to optimize its understanding model of user intent.

[0057] The system will iterate in this closed loop, with each iteration based on the previous in-depth diagnosis, thus intelligently approaching the optimal solution until the "evaluation report" shows that all dimensions meet the requirements, and finally outputs high-quality results.

[0058] Through iterative optimization of blueprint generation, the system can feed back the experience of each abductive learning session to the initial semantic understanding stage, continuously optimizing its ability to deconstruct user intent and formulate generation blueprints. This enables the system to have continuous learning capabilities, and its understanding of user intent will become more and more accurate with the increase in the number of interactions, thereby reducing the possibility of generation failure from the root.

[0059] In this embodiment, an intelligent generation workflow system with autonomous cognitive feedback and abductive reasoning capabilities is achieved through cognitive detection, abductive reasoning, and dynamic adjustment of the workflow path structure. This enables the workflow to not only generate content but also detect defects in the output content, deeply deduce the root causes of defects, and autonomously formulate and execute optimization strategies based on these defects, i.e., formulate precise strategies for improvement. This system breaks the traditional linear pipeline generation mode and establishes a dynamic intelligent system with a cognitive closed loop as its core, thereby realizing a paradigm shift in the system's autonomous iterative optimization. It achieves refined and intelligent management of the entire generation workflow process, significantly improving generation quality and efficiency, ultimately achieving a dual improvement in generation efficiency and quality, while also reducing the professional technical requirements for users.

[0060] By employing an intelligent generation workflow optimization method based on cognitive feedback and semantic abduction, this system introduces semantic abduction and cognitive feedback mechanisms to create an AI generation system with metacognitive capabilities. It is no longer a passive tool but an active, collaborative intelligent agent, endowing the workflow with unprecedented depth of understanding and decision-making. This achieves a leap in automation and intelligence in the generation process, significantly reducing human intervention and increasing the efficiency of high-quality content generation by orders of magnitude. Secondly, the system possesses expert-level problem diagnosis and resolution capabilities, producing content with more rigorous logic, more accurate semantics, and higher consistency. Finally, the system greatly lowers the user threshold; ordinary users do not need to master complex prompting engineering skills, but can obtain professional-level generated output simply through natural language descriptions, greatly promoting the popularization and application depth of AI generation technology. Through a next-generation AI generation workflow system with self-awareness, self-diagnosis, and self-optimization capabilities, its core idea is to mimic the creative process of human experts: not only generating content but also critically examining it, understanding its shortcomings, and formulating precise strategies for improvement. This system breaks the traditional "linear pipeline" generation model and establishes a dynamic intelligent system centered on a "cognitive closed loop."

[0061] Figure 3 A schematic diagram of a workflow optimization device based on cognitive feedback and semantic abduction is provided. Figure 3 As shown, the workflow optimization device 300 based on cognitive feedback and semantic abduction includes: The generation module 301 is used to generate initial requirement feedback content in response to an input operation for user needs, based on the target user needs corresponding to the input operation. The detection module 302 is used to detect the initial demand feedback content using a preset cognitive detection model and generate cognitive defect data of the initial demand feedback content. The tracing module 303 is used to reverse deduce the generation of the cognitive defect data through the tracing reasoning engine, obtain the semantic tracing result of the cognitive defect data, and map the cognitive defect data back to the workflow path of the initial demand feedback content in the generation process. The reconstruction module 304 is used to dynamically adjust the workflow path according to the semantic tracing result, and iteratively reconstruct the requirement feedback content corresponding to the target user requirement based on the adjusted workflow path to obtain the final requirement feedback content.

[0062] The workflow optimization device based on cognitive feedback and semantic abduction provided in this application has the same technical features as the workflow optimization method based on cognitive feedback and semantic abduction provided in the above embodiments, so it can also solve the same technical problems and achieve the same technical effects.

[0063] An electronic device provided in this application embodiment, such as Figure 4 As shown, the electronic device 400 includes a processor 402 and a memory 401. The memory stores a computer program that can run on the processor. When the processor executes the computer program, it implements the steps of the method provided in the above embodiments.

[0064] See Figure 4 The electronic device also includes a bus 403 and a communication interface 404. The processor 402, the communication interface 404 and the memory 401 are connected via the bus 403. The processor 402 is used to execute executable modules, such as computer programs, stored in the memory 401.

[0065] The memory 401 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 404 (which can be wired or wireless), such as the Internet, wide area network, local area network, or metropolitan area network.

[0066] Bus 403 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 4 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0067] The memory 401 is used to store programs. After receiving an execution instruction, the processor 402 executes the program. The method executed by the apparatus defined by the process disclosed in any of the preceding embodiments of this application can be applied to the processor 402 or implemented by the processor 402.

[0068] Processor 402 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 402 or by instructions in software form. The processor 402 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software module can reside in a mature storage medium in the field, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory 401, and processor 402 reads the information from memory 401 and, in conjunction with its hardware, completes the steps of the above method.

[0069] Corresponding to the above-described workflow optimization method based on cognitive feedback and semantic abduction, this application embodiment also provides a computer-readable storage medium storing computer-executable instructions. When the computer-executable instructions are invoked and executed by a processor, the computer-executable instructions cause the processor to perform the steps of the above-described workflow optimization method based on cognitive feedback and semantic abduction.

[0070] The workflow optimization device based on cognitive feedback and semantic reasoning provided in this application embodiment can be specific hardware on a device or software or firmware installed on the device. The implementation principle and technical effects of the device provided in this application embodiment are the same as those in the foregoing method embodiments. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the foregoing method embodiments. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can all be referred to the corresponding processes in the above method embodiments, and will not be repeated here.

[0071] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0072] For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked 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 a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can 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.

[0073] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0074] In addition, the functional units in the embodiments provided in this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0075] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the workflow optimization method based on cognitive feedback and semantic attribution described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0076] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0077] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The protection scope of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application. All should be covered within the protection scope of this application. Therefore, the protection scope of this application should be determined by the protection scope of the claims.

Claims

1. A workflow optimization method based on cognitive feedback and semantic abduction, characterized in that, The method includes: In response to an input operation that addresses a user's needs, an initial feedback content is generated based on the target user's needs corresponding to the input operation. Based on the initial demand feedback content, a preset cognitive detection model is used to detect and generate cognitive defect data of the initial demand feedback content; The cognitive defect data is reverse-engineered through the abductive reasoning engine to obtain the semantic abductive result of the cognitive defect data, and the cognitive defect data is mapped back to the workflow of the initial demand feedback content in the generation process. The workflow path structure is dynamically adjusted based on the semantic causation results. The requirement feedback content corresponding to the target user's requirement is iteratively reconstructed based on the adjusted workflow path structure to obtain the final requirement feedback content. The step of responding to an input operation targeting a user's needs and generating initial demand feedback content based on the target user's needs corresponding to the input operation includes: responding to an input operation targeting a user's needs, performing deep semantic understanding and user demand task parsing on the natural language description content corresponding to the input operation through an integrated large-scale language model to obtain the deep intent, core theme, sentiment tendency, style requirements, implicit constraints, and key entities of the task corresponding to the user's needs; generating structured initial demand feedback blueprint data based on the deep intent, the core theme, the sentiment tendency, the style requirements, the implicit constraints, and the key entities; and converting the initial demand feedback blueprint data into refined instructions through a multimodal generation module, and driving the corresponding text model, image model, and / or audio model to generate initial demand feedback content through the instructions. The step of detecting the initial demand feedback content using a preset cognitive detection model to generate cognitive defect data for the initial demand feedback content includes: verifying the accuracy and consistency of factual statements in the initial demand feedback content using a search engine based on the knowledge graph corresponding to the preset cognitive detection model; in response to the factual statements meeting preset accuracy and consistency requirements, performing logical coherence analysis on the argumentation structure and causal chain of the initial demand feedback content using the preset cognitive detection model to determine whether logical fallacies exist; in response to the absence of logical fallacies, determining whether the emotional tone and style of the initial demand feedback content conform to the initial demand feedback blueprint data; in response to conforming to the initial demand feedback blueprint data, detecting whether the semantic matching degree of the initial demand feedback content meets a preset matching degree through cross-modal consistency verification; and in response to the semantic matching degree not meeting the preset matching degree, generating cognitive defect data for the initial demand feedback content in the semantic matching dimension. The cognitive defect data refers to the factual statements in the initial demand feedback content not meeting the preset accuracy requirements. The step of reverse-deriving the generation of the cognitive defect data through the abductive reasoning engine to obtain the semantic abductive result of the cognitive defect data includes: reverse-deriving the generation of the cognitive defect data through the abductive reasoning engine to determine that the root cause of the semantic abductive result of the cognitive defect data is that the contextual definition of key entities in the initial demand feedback content is ambiguous, which leads to the generation model activating incorrect knowledge associations.

2. The method according to claim 1, characterized in that, The process of detecting cognitive deficiencies in the initial demand feedback content using a preset cognitive detection model, and generating cognitive defect data for the initial demand feedback content, includes: Based on the user intent corresponding to the target user's needs, retrieve all relevant data and determine whether the target user's needs are original based on the quantity of the relevant data and its relevance to the user intent. Since the target user's need is not an original need, the main learning time period of the preset cognitive detection model is determined based on the generation time of the relevant data; different learning time periods correspond to different cognitive content; Based on the target cognitive content corresponding to the main learning time period, the preset cognitive detection model is cognitively optimized to obtain the first target cognitive detection model. Based on the initial demand feedback content, the first target cognitive detection model is used to detect and generate cognitive defect data of the initial demand feedback content.

3. The method according to claim 2, characterized in that, After determining whether the target user's need is an original need based on the quantity of the relevant data and its relevance to the user's intent, the method further includes: Since the target user's need is an original need, the target user type is identified based on the natural language description content corresponding to the input operation and the user intent corresponding to the target user's need; the user type includes at least one of user age, user industry field, user occupation category, and user professional direction; different user types correspond to different cognitive directions and business logics. Based on the target business logic and target cognitive direction corresponding to the target user type, the preset cognitive detection model is optimized in a logical and cognitive manner to obtain a second target cognitive detection model. Based on the initial demand feedback content, the second target cognitive detection model is used to detect and generate cognitive defect data of the initial demand feedback content.

4. The method according to claim 1, characterized in that, The process involves dynamically adjusting the workflow path structure based on the semantic causation results, iteratively reconstructing the requirement feedback content corresponding to the target user's needs based on the adjusted workflow path structure, and obtaining the final requirement feedback content, including: Based on the semantic abduction results, structured decisions are generated through the workflow scheduling center, and the semantic abduction results are transformed into optimization instructions; In response to the detection that the semantic causation result contains a knowledge blind spot, the knowledge retrieval and integration sub-process node corresponding to the structured decision is inserted into the workflow based on the optimization instruction to obtain the path structure of the adjusted workflow; In response to the detection that the semantic causation result contains an error in the deconstruction of the user's demand task, the structured decision is fed back to the semantic understanding sub-process node in the workflow based on the optimization instruction, so as to reconstruct and optimize the deconstruction model of the user intent in the semantic understanding sub-process node. The above steps are executed repeatedly, and in each iteration, the workflow path structure is iterated based on the cognitive deficit data and the semantic causal results, until there is no cognitive deficit data in all dimensions of the requirement feedback content, and the final requirement feedback content is output.

5. A workflow optimization device based on cognitive feedback and semantic abduction, characterized in that, include: The generation module is used to respond to input operations that address user needs and generate initial feedback content based on the target user needs corresponding to the input operations. The detection module is used to detect the initial demand feedback content using a preset cognitive detection model and generate cognitive defect data of the initial demand feedback content. The causal module is used to reverse deduce the generation of the cognitive defect data through the causal reasoning engine, obtain the semantic causal result of the cognitive defect data, and map the cognitive defect data back to the workflow path of the initial demand feedback content in the generation process. The reconstruction module is used to dynamically adjust the workflow path based on the semantic causation results, and iteratively reconstruct the requirement feedback content corresponding to the target user requirement based on the adjusted workflow path to obtain the final requirement feedback content. The generation module is specifically used for: responding to input operations based on user needs, performing deep semantic understanding and user need task parsing on the natural language description content corresponding to the input operation through an integrated large-scale language model, to obtain the deep intent, core theme, sentiment tendency, style requirements, implicit constraints, and key entities of the user need task; generating structured initial need feedback blueprint data based on the deep intent, core theme, sentiment tendency, style requirements, implicit constraints, and key entities; and converting the initial need feedback blueprint data into refined instructions through a multimodal generation module, and driving the corresponding text model, image model, and / or audio model to generate initial need feedback content through the instructions. The detection module is specifically used to: verify the accuracy and consistency of factual statements in the initial demand feedback content using a search engine based on the knowledge graph corresponding to the preset cognitive detection model; in response to the factual statements meeting the preset accuracy and consistency requirements, perform logical coherence analysis on the argumentation structure and causal chain of the initial demand feedback content using the preset cognitive detection model to determine whether there are any logical fallacies. In response to the absence of logical fallacies, it is determined whether the emotional tone and style of the initial demand feedback content conform to the initial demand feedback blueprint data; in response to conforming to the initial demand feedback blueprint data, it is detected whether the semantic matching degree of the initial demand feedback content conforms to the preset matching degree through cross-modal consistency verification. In response to the semantic matching degree not meeting the preset matching degree, cognitive defect data of the initial demand feedback content in the semantic matching dimension is generated; The cognitive defect data refers to the factual statements in the initial demand feedback content not meeting the preset accuracy requirements; the causal module is specifically used to: reverse-derive the generation of the cognitive defect data through the causal reasoning engine, and determine that the root cause of the semantic causal result of the cognitive defect data is that the contextual definition of key entities in the initial demand feedback content is ambiguous, which leads to the generation model activating incorrect knowledge associations.

6. An electronic device comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions that, when invoked and executed by a processor, cause the processor to perform the method according to any one of claims 1 to 4.

Images