Adaptive task decomposition method and system for achievement transformation large model

Abstract

The invention belongs to the technical field of large model application, and particularly relates to an achievement transformation large model-oriented adaptive task decomposition method and system. The self-adaptive task decomposition method oriented to the large achievement transformation model comprises five parts, namely analysis of task receiving and initialization, formulation of a task decomposition strategy, execution of task decomposition, self-adaptive optimization of task decomposition and summarization and updating of experience and knowledge. And task receiving and initialization analysis is used for carrying out structured understanding and complexity evaluation on an input task and providing a basic analysis basis for subsequent task decomposition. Formulating a task decomposition strategy is used for determining decomposition parameters and decomposition templates suitable for the current task, and providing specific strategy guidance for task decomposition execution. The execution of the task decomposition is used for decomposing the complex task into sub-tasks meeting the granularity requirement and optimizing the dependency relationship among the sub-tasks. Adaptive optimization of task decomposition is used for evaluating and optimizing a decomposition result, and it is ensured that the decomposition result meets actual requirements through multiple rounds of optimization. Summarizing and updating of experience and knowledge are used for summarizing task decomposition experience and updating a knowledge base, and experience support is provided for subsequent task decomposition. Through systematized task evaluation, a dynamic self-adaptive decomposition strategy and a perfect knowledge accumulation mechanism, complex requirements in the achievement conversion process can be effectively processed, and the method is particularly suitable for actual conversion scenes such as patent technology landing and scientific achievement industrialization.

Description

Background Art

[0002] With the rapid development of large language model technology, transforming research results into practical applications has become an important trend. In the process of transformation, users often need to put forward complex transformation requirements to large language models, such as the implementation of patented technologies and the industrialization of scientific research results. However, such transformation requirements usually involve multiple professional fields, contain complex implementation steps, and require detailed reasoning and argumentation processes, which makes it difficult for the model to provide a complete and practical transformation plan at one time. Therefore, decomposing complex results transformation tasks into actionable subtasks has become one of the key technologies to improve the application effect of large language models.

[0003] At present, the task decomposition technology for the achievement transformation scenario mainly includes task decomposition based on prompt engineering and task decomposition based on rules. Task decomposition based on prompt engineering guides the model to decompose complex transformation tasks into specific implementation steps by designing special prompt templates. Rule-based task decomposition predefines a series of transformation rules and patterns, and selects the appropriate decomposition method according to the characteristics of the results. However, the inventors found that the existing task decomposition technology has obvious shortcomings in the achievement transformation scenario. First, the lack of a systematic assessment of the complexity of the transformation task leads to unreasonable granularity of the decomposed implementation steps. Secondly, the decomposition strategy is too fixed and cannot be dynamically adjusted according to the actual feedback in the transformation process. In addition, the capabilities of the large language model are not fully utilized to optimize the transformation plan. Finally, due to the lack of a complete experience accumulation and knowledge updating mechanism, it is difficult to continuously improve the effect of achievement transformation. Summary of the invention

[0004] In order to solve at least one of the technical problems existing in the above-mentioned background technology, the present invention provides an adaptive task decomposition method and system for a large model of achievement transformation, which can significantly improve the decomposition quality of complex tasks, thereby improving the accuracy and completeness of the answers of the large language model, and providing better services for users.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] The first aspect of the present invention provides an adaptive task decomposition method for a large model of achievement transformation, which includes:

[0007] Analysis of task reception and initialization. Preprocess the input task text, including text cleaning, word segmentation, and entity recognition, and then perform complexity assessment. Use dependency syntactic analysis to build a dependency graph and calculate the complexity score. At the same time, calculate the knowledge domain range score. Finally, perform complexity analysis, normalize the scores of the two dimensions, and map them to three levels: low / medium / high.

[0008] Formulation of task decomposition strategy. Query the historical experience knowledge base, extract task features and retrieve similar cases, then determine the task decomposition parameters, including determining the number of subtasks based on the complexity level and setting the task granularity requirements, and finally select the appropriate task decomposition template and make customized adjustments based on the parameters.

[0009] Execution of task decomposition. Perform initial decomposition, generate prompt words based on the selected template and obtain the initial decomposition results, then perform detailed decomposition, check and optimize subtasks that do not meet the granularity requirements, and finally optimize the dependencies between tasks to ensure the rationality of the decomposition results.

[0010] Adaptive optimization of task decomposition. Submit subtasks to the language model and evaluate the quality of the answer, then identify existing problems and analyze the causes, generate targeted optimization strategies, and finally perform optimization and verify the results, and perform multiple rounds of optimization if necessary.

[0011] Update the summary of experience and knowledge. Update the task decomposition template library, including template information, parameter configuration and usage records, update the optimization strategy library, including problem types, optimization solutions and effect evaluation, and update the case library, including complete task decomposition records and experience summaries.

[0012] The second aspect of the present invention provides an adaptive task decomposition system for a large model of achievement transformation. It includes:

[0013] The task receiving and initialization analysis module forms a structured understanding of the task through task text preprocessing and complexity assessment, including task dependencies, knowledge domain coverage, and ultimately determines the complexity level of the task, providing a basic analysis basis for subsequent task decomposition;

[0014] The task decomposition strategy formulation module combines the historical experience knowledge base to determine the decomposition parameters and decomposition templates suitable for the current task, including determining the decomposition level, task granularity requirements, and selecting appropriate decomposition templates, providing specific strategic guidance for task decomposition execution;

[0015] The task decomposition execution module performs initial decomposition, detailed decomposition and dependency optimization of tasks according to the formulated decomposition strategy to ensure that each subtask meets the granularity requirements and forms a reasonable task decomposition structure;

[0016] The task decomposition adaptive optimization module uses the existing mature large language model to evaluate the quality of the decomposition results, identify existing problems, formulate and implement optimization strategies, and improve the effect of task decomposition through multiple rounds of optimization to ensure that the decomposition results meet actual needs;

[0017] The experience and knowledge summary and update module summarizes the experience of the entire process of this task decomposition and updates it to the knowledge base, including updating the task decomposition template library, optimization strategy library and case library, to provide experience support and knowledge accumulation for subsequent task decomposition.

[0018] A third aspect of the present invention provides a computer-readable storage medium.

[0019] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps in the adaptive task decomposition method for a large model of achievement transformation as described above.

[0020] A fourth aspect of the present invention provides a computer device.

[0021] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the adaptive task decomposition method for a large model of achievement transformation as described above are implemented.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] (1) The adaptive task decomposition method for the large model of achievement transformation proposed in this invention can significantly improve the decomposition quality of complex tasks, thereby improving the accuracy and completeness of the answers of the large language model and providing better services for users.

[0024] (2) The present invention establishes a systematic task complexity evaluation system. The system evaluates tasks from two dimensions: dependency and knowledge domain. In the dependency dimension, the task dependency graph is constructed through dependency syntax analysis, and the complexity score is calculated based on the longest path of the graph. In the knowledge domain dimension, the domain coverage is calculated through vectorized representation and knowledge graph matching.

[0025] (3) The present invention implements a dynamic optimization mechanism based on feedback. This method is different from the traditional static decomposition method. Instead, it designs a complete feedback evaluation and optimization process. It uses a large language model to evaluate the decomposition results in multiple dimensions, including completeness, accuracy, and relevance. Then, it identifies problems and generates optimization strategies based on the evaluation results. Finally, it continuously improves the decomposition quality through multiple rounds of optimization.

[0026] (4) The present invention constructs a complete knowledge accumulation and updating system and designs three core knowledge bases: the task decomposition template library stores various decomposition patterns and usage experience, the optimization strategy library accumulates problem solutions, and the case library precipitates complete decomposition cases.

[0027] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0029] Figure 1 It is a flow chart of an adaptive task decomposition method for a large model of achievement transformation according to an embodiment of the present invention;

[0030] Figure 2 is a flowchart of task reception and initialization analysis of an embodiment of the present invention;

[0031] Figure 3 is a flow chart of task decomposition strategy formulation according to an embodiment of the present invention;

[0032] Figure 4 is a flowchart of task decomposition and execution of an embodiment of the present invention;

[0033] Figure 5 is a flowchart of task decomposition adaptive optimization according to an embodiment of the present invention;

[0034] Figure 6 It is an experience and knowledge summary and update flow chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0035] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0036] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0037] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0038] Embodiment 1

[0039] Reference Figure 1 This embodiment provides an adaptive task decomposition method for a large model of achievement transformation, which includes:

[0040] S101: Analysis of task reception and initialization, pre-processing of the description of the task input by the user, and forming a preliminary level of task complexity.

[0041] S102: Formulate task decomposition strategy by querying the historical experience database to match similar cases and select appropriate decomposition templates and rules.

[0042] S103: Execution of task decomposition: performing initial decomposition based on the selected template, refining the decomposition results, and optimizing the dependencies between subtasks.

[0043] S104: Adaptive optimization of task decomposition, evaluate decomposition effect, identify problems, formulate optimization strategy, execute optimization and verify the effect.

[0044] S105: Summarize and update experience and knowledge, and update the task decomposition template library, optimization strategy library and case library.

[0045] Reference Figure 2 , this embodiment provides a step flow chart of task reception and initialization analysis, and describes the analysis of task reception and initialization in detail.

[0046] Step 200: Preprocessing of task description. Preprocess the input task text (cleaning, word segmentation, entity recognition). Step 200 can be further divided into the following steps:

[0047] Step 200-1: Enter task description;

[0048] Step 200-2: Use regular expressions to clean special characters and redundant spaces in the text, and then use a word segmentation model (such as BERT) to segment the text into word sequences;

[0049] Step 200 - 3 : For the word sequence in step 200 - 1 , use part-of-speech tagging to identify key verbs, nouns, modifiers, etc. in the text, and extract entity information such as specific objects, time, and place in the task through named entity recognition.

[0050] Step 201: Evaluate the complexity of the task description. Evaluate the complexity of the task description from the dependency relationship and knowledge domain scope. Step 201 can be further refined into the following steps:

[0051] Step 201-1: for the task description preprocessed in step 200, extract the verb phrases and subject-predicate relationships in the task text using dependency syntactic analysis, and construct a dependency graph, where each node represents a subtask or entity, and the edge represents a dependency relationship;

[0052] Step 201-2: Calculate the dependency complexity score by using a graph analysis algorithm, where the dependency is measured by the longest path length;

[0053] Step 201-3: Record the node information of the longest path in the dependency relationship;

[0054] Step 201-4: convert the noun text identified in step 200-2 into a vector representation (pre-trained models such as BERT or Word2Vec can be used), and then calculate the cosine similarity between these vectors and the entity vectors in the knowledge graph, and count the number of knowledge graph entities whose similarity is lower than a preset threshold (such as 0.3). This number can represent the score of the knowledge domain involved in the task. The knowledge graph can use the public Microsoft Academic Knowledge Graph, etc.;

[0055] Step 201 - 5 : Record the entity names involved in the knowledge domain and the corresponding cosine similarities.

[0056] Step 202: Analysis of task complexity. Step 202 can be further broken down into the following steps:

[0057] Step 202-1: using the dependency score obtained in step 201-3 and the knowledge domain range score obtained in step 201-5, normalize them to the range of 0-1 through min-max;

[0058] Step 202-2: Obtain the complexity score of the computing task using weighted average;

[0059] Step 202 - 3 : Map the scores to three levels of low / medium / high based on preset thresholds (eg, 0.4 and 0.7).

[0060] Reference Figure 3 This embodiment provides a flowchart of the steps for formulating a task decomposition strategy, which describes the formulation of the task decomposition strategy in detail, including:

[0061] Step 300: Query the historical experience knowledge base. Query and analyze the existing experience knowledge. Step 300 can be further divided into the following steps:

[0062] Step 300-1: extracting key features of the task (such as the knowledge domain scope and dependency relationships recorded in step 200);

[0063] Step 300-2: construct a query vector based on the key features and retrieve similar task cases in the experience database;

[0064] Step 300-3: Calculate case similarity scores and filter out cases with similarities higher than a preset threshold;

[0065] Step 300-4: Analyze the selected cases and summarize reusable experience patterns.

[0066] Step 301: Determine the task decomposition parameters. Determine the decomposition parameters based on historical experience and current task characteristics. The task decomposition parameters include the number of subtasks in the task decomposition and the complexity of the dependency relationship of each subtask. Step 301 can be further refined into the following steps:

[0067] Step 301-1: Determine the decomposition level according to the complexity level obtained in step 202-3 (1-2 subtasks for low complexity, 2-3 subtasks for medium complexity, and 3-4 subtasks for high complexity), which is the requirement for the number of subtasks;

[0068] Step 301-2: Based on the dependency node information recorded in step 201-4, set the task granularity range of each subtask. The dependency complexity score of each subtask should be less than β% of the dependency complexity score of the original task (can be set by the user). This is the task granularity requirement.

[0069] Step 301 - 3: Optimize and adjust the parameters in combination with the experience model of step 300 - 4;

[0070] Step 301 - 4 : Output the final task decomposition parameter solution.

[0071] Step 302: Selection of a task decomposition template. Select an appropriate task decomposition template. Step 302 can be further divided into the following steps:

[0072] Step 302-1: Collect and organize different standard work breakdown structure templates to build an initial decomposition template library, such as a sequential decomposition template, a parallel decomposition template, etc.;

[0073] Step 302-2: Customize the selected template according to the parameter scheme of step 301-4;

[0074] Step 302-3: Determine the task decomposition template solution to be finally used.

[0075] Reference Figure 4 This embodiment provides a flowchart of the steps of task decomposition execution, which describes the execution of task decomposition in detail, including:

[0076] Step 400: Initialization of task decomposition. Perform a preliminary decomposition of the task. Step 400 can be further broken down into the following steps:

[0077] Step 400-1: Obtain the task decomposition template solution determined in step 302-4;

[0078] Step 400-2: Generate an initial prompt word based on the task decomposition template. For example, the basic prompt word of the sequential decomposition template may be "Please decompose the following task into {number of subtasks} consecutive steps {original task} in chronological order, and the requirement is that each step should be a natural continuation of the previous step, and the dependency relationship between the steps should be clearly pointed out. The granularity of each step should meet {task granularity requirements}";

[0079] Step 400 - 3: The number of subtasks and task granularity in step 400 - 3 are obtained from step 301 - 4 , and the task decomposition parameters are filled into the prompt word;

[0080] Step 400-4: Send the filled prompt words to the language model;

[0081] Step 400 - 5 : Receive and store the initial decomposition result returned by the language model.

[0082] Step 401: Refine the task. Refine the initial decomposition result. Step 401 can be further refined into the following steps:

[0083] Step 401-1: Analyze the initial decomposition result obtained in step 400-6;

[0084] Step 401 - 2: Check whether each subtask meets the task granularity requirement set in step 301 - 2;

[0085] Step 401-3: For subtasks that do not meet the granularity requirement, analyze the reasons for not meeting the granularity requirement, and generate new prompt words for further decomposition. For example, if it is analyzed that the subtask is too large (the complexity exceeds the threshold, the complexity calculation is as shown in step 202-2, and the threshold is set in advance), the new prompt words may be "Please further split the following subtasks into smaller steps: {subtask description}, and the requirement is that the complexity of each step should not exceed {complexity threshold}, maintain logical coherence between steps, and clearly state the specific goals of each step";

[0086] Step 401-4: Send the new prompt word to the language model;

[0087] Step 401 - 5 : Receive and integrate the refined decomposition results returned by the language model.

[0088] Step 402: Optimize task dependencies. Optimize dependencies between subtasks. Step 402 can be further broken down into the following steps:

[0089] Step 402-1: Analyze the dependency relationship between subtasks based on the dependency node information recorded in step 201-4;

[0090] Step 402-2: Check the rationality of the dependency relationship and identify possible circular dependencies or redundant dependencies;

[0091] Step 402-3: Optimize and adjust unreasonable dependencies;

[0092] Step 402-4: Update and save the final task decomposition result.

[0093] Reference Figure 5 This embodiment provides a flowchart of the steps of task decomposition adaptive optimization, which describes the adaptive optimization of task decomposition in detail, including:

[0094] Step 500: Evaluation of the effect of task decomposition. Evaluate the result of task decomposition. Step 500 can be further divided into the following steps:

[0095] Step 500-1: Obtain the task decomposition result saved in step 402-4;

[0096] Step 500-2: Submit each decomposed subtask to the language model to obtain the answer result;

[0097] Step 500-3: Evaluate the answer quality of each subtask, including completeness, accuracy, and relevance. To evaluate the answer quality of subtasks, large models such as GPT-4 and Claude can be used as evaluators.

[0098] Step 501: Problem identification and analysis. Identify problems in task decomposition. This process can also use large models such as GPT-4 and Claude as analyzers. Analyze the reasons for low answer quality, which may include: unclear task description, inappropriate task granularity, incorrect dependencies, etc., and generate a problem analysis report.

[0099] Step 502: Generate an optimization strategy. Generate a targeted optimization strategy. Step 502 can be further divided into the following steps:

[0100] Step 502-1: Based on the problem analysis report of step 501, formulate an optimization strategy for each type of problem;

[0101] Step 502-2: query and analyze the optimization records of similar tasks in the historical experience database to extract reusable optimization patterns;

[0102] Step 502-3: Customize the optimization plan based on the characteristics of the current task, for example, select the corresponding optimization strategy according to the problem type (unclear subtask description, inappropriate task granularity, incorrect dependency), and determine the key direction and priority of optimization based on the evaluation results (completeness, accuracy, relevance);

[0103] Step 502-4: Generate an optimization strategy execution plan.

[0104] Step 503: Execution and verification of optimization. Execute the optimization strategy and verify the effect. Step 503 can be further divided into the following steps:

[0105] Step 503-1: Implement optimization according to the optimization strategy execution plan of step 502-4;

[0106] Step 503-2: re-execute the evaluation process of step 500 on the optimized decomposition result;

[0107] Step 503-3: Compare the evaluation scores before and after optimization;

[0108] Step 503-4: If the optimization effect is not satisfactory, return to step 502 to formulate a new optimization strategy.

[0109] Reference Figure 6 This embodiment provides a flowchart of the steps of summarizing and updating experience and knowledge, and describes in detail the summary and update of experience and knowledge, which includes:

[0110] Step 600: Update of the task decomposition template library. The core storage structure includes: basic information (template ID, template type such as the sequential decomposition template and parallel decomposition template defined in step 302-1, etc.), prompt word information (based on the basic prompt word template generated in step 400-3, such as "Please decompose the following task into {number of subtasks} consecutive steps in chronological order..."), parameter configuration (from steps 301-1 and 301-2, including the number of subtasks based on complexity level and the task granularity based on dependency), usage constraints (based on the dependency complexity of step 201-3 and the dependency node information of 201-4), domain applicability conditions (based on the knowledge domain scope score of step 201-5 and the knowledge domain entity information of 201-6), evaluation records (including the task ID using the template, the evaluation score of step 500, the problem analysis record of step 501, and the optimization record of step 502), optimization suggestions (based on the optimization experience summarized by the verification result of step 503) and associated template recommendations;

[0111] Step 601: Update of optimization strategy library. The core storage structure includes: problem type (based on the problem type analyzed in step 501, such as unclear task description, inappropriate task granularity, incorrect dependency, etc.), evaluation dimension (based on the evaluation dimension of step 500-3, including completeness, accuracy, and relevance), optimization strategy (based on the specific optimization plan formulated for each type of problem in step 502-1), strategy effect (through the comparison of evaluation scores before and after optimization recorded in steps 503-2 and 503-3), usage constraints (based on the task feature analysis in step 502-3, including the circumstances under which the strategy is applicable), implementation plan (optimization strategy execution plan generated in step 502-4), associated tasks (recording the task ID using the optimization strategy), and optimization experience (based on the multi-round optimization records in step 503-4, including strategy adjustment history and effect analysis);

[0112] Step 602: Update of the case library. The core storage structure includes: basic task information (preprocessing results of step 200, including keywords, entity information, etc.), task complexity characteristics (analysis results of steps 201 and 202, including dependency complexity, knowledge domain range score, and final complexity level), decomposition strategy records (decomposition parameter scheme of step 301 and template selection results of step 302), execution process information (initial decomposition results of step 400, refinement process of step 401, dependency optimization record of step 402), evaluation and optimization records (evaluation results of step 500, problem analysis of step 501, optimization strategy of step 502, optimization effect of step 503), failure lessons (recording the reasons for optimization failure and improvement suggestions), successful experience (recording the key success factors of high-scoring cases), and associations with other similar cases (based on the similarity calculation results of steps 300-2 and 300-3).

[0113] Embodiment 2

[0114] This embodiment provides an adaptive task decomposition system for a large model of achievement transformation. It includes:

[0115] The task receiving and initialization analysis module forms a structured understanding of the task through task text preprocessing and complexity assessment, including task dependencies, knowledge domain coverage, and ultimately determines the complexity level of the task, providing a basic analysis basis for subsequent task decomposition;

[0116] The task decomposition strategy formulation module combines the historical experience knowledge base to determine the decomposition parameters and decomposition templates suitable for the current task, including determining the decomposition level, task granularity requirements, and selecting appropriate decomposition templates, providing specific strategic guidance for task decomposition execution;

[0117] The task decomposition execution module performs initial decomposition, detailed decomposition and dependency optimization of tasks according to the formulated decomposition strategy to ensure that each subtask meets the granularity requirements and forms a reasonable task decomposition structure;

[0118] The task decomposition adaptive optimization module uses the existing mature large language model to evaluate the quality of the decomposition results, identify existing problems, formulate and implement optimization strategies, and improve the effect of task decomposition through multiple rounds of optimization to ensure that the decomposition results meet actual needs;

[0119] The experience and knowledge summary and update module summarizes the experience of the entire process of this task decomposition and updates it to the knowledge base, including updating the task decomposition template library, optimization strategy library and case library, to provide experience support and knowledge accumulation for subsequent task decomposition.

[0120] Embodiment 3

[0121] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps in the adaptive task decomposition method for the large model of achievement transformation as described above are implemented.

[0122] Embodiment 4

[0123] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the adaptive task decomposition method for the large model of achievement transformation as described above are implemented.

[0124] It should be appreciated by those skilled in the art that embodiments of the present invention may be in the form of methods, systems, or computer program products. Therefore, the invention may be implemented in the form of hardware, software, or a combination of hardware and software. In addition, the invention may be in the form of a computer program product on a computer-usable storage medium (including but not limited to a disk storage and an optical storage, etc.), which contains one or more computer-usable program codes.

[0125] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An adaptive task decomposition technology for large-scale achievement transformation models, characterized in that: It includes the analysis of task reception and initialization, the formulation of task decomposition strategy, the execution of task decomposition, the adaptive optimization of task decomposition, and the summary and update of experience and knowledge. Among them, the analysis of task reception and initialization is used to structurally understand and evaluate the complexity of the input task. The formulation of task decomposition strategy is used to determine the decomposition parameters and decomposition template suitable for the current task. The execution of task decomposition is used to decompose complex tasks into subtasks that meet the granularity requirements and optimize the dependencies between subtasks. The adaptive optimization of task decomposition is used to evaluate and optimize the decomposition results, and ensure that the decomposition results meet the actual needs through multiple rounds of optimization. The summary and update of experience and knowledge is used to summarize the task decomposition experience and update the knowledge base.

2. An adaptive task decomposition technology for a large model of achievement transformation is used to analyze the process of task reception and initialization, characterized in that: The task text is preprocessed to identify key information through text cleaning, word segmentation and part-of-speech tagging. Named entity recognition is used to extract specific entity information. Then, a dependency graph is constructed through dependency syntactic analysis. The dependency complexity score based on the longest path length is calculated, and key node information is recorded.

3. An adaptive task decomposition technology for a large model of achievement transformation to analyze the process of task reception and initialization, characterized in that: After the nominal text is vectorized, the similarity is calculated with the knowledge graph entity, and the number of entities with similarity below the threshold is counted to obtain the knowledge domain scope score. According to claim 2, the dependency complexity score and the knowledge domain scope score are normalized and weighted averaged to obtain the final task complexity level.

4. A process of formulating a task decomposition strategy using an adaptive task decomposition technology for a large model of achievement transformation, characterized in that: According to claim 3, the historical experience database is queried to obtain similar cases, the decomposition parameters are determined in combination with the complexity of the current task, and the appropriate decomposition template is selected and adjusted.

5. An adaptive task decomposition technology for a large model of achievement transformation performs a task decomposition execution process, characterized in that: Perform initial decomposition according to the template selected in claim 4, refine the decomposition result, and optimize the dependencies between subtasks.

6. An adaptive task decomposition technology for a large model of achievement transformation performs an adaptive optimization process of task decomposition, characterized by Submit the decomposed subtasks to the language model and evaluate the completeness, accuracy, and relevance of the answer.

7. An adaptive task decomposition technology for a large model of achievement transformation performs an adaptive optimization process of task decomposition, characterized by According to claim 6, an optimization strategy is formulated based on problem analysis and historical experience, the optimization direction and priority are determined, the optimization is executed and the effect is verified, and if necessary, multiple rounds of optimization are performed until the expected effect is achieved.

8. An adaptive task decomposition technology for the large-scale model of achievement transformation to summarize and update experience and knowledge, characterized by According to claims 1-7, the task decomposition template library, the optimization strategy library and the case library are updated.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the adaptive task decomposition method for a large model for achievement transformation as described in any one of claims 1 to 7 are implemented.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, it implements the steps in the adaptive task decomposition method for a large model of achievement transformation as described in any one of claims 1-7.

Images