A method for human-machine hierarchical collaborative decision-making and contribution right determination in multidisciplinary education and scientific research

Abstract

The application discloses a kind of multidisciplinary education scientific research man-machine layered collaborative decision-making and contribution right method, comprising the following steps: S1: multidisciplinary teaching research task standardization disassembly stage.S2: state perception man-machine layered collaborative decision-making execution stage.S3: multidimensional man-machine innovation contribution layered quantification stage.S4: whole-process data encryption storage and automatic right allocation stage.The application solves the defects of rigid allocation of prior art permissions, subjective and one-sided contribution quantification, non-traceable decision-making process and single evaluation dimension.

Description

Technical Field

[0001] This invention relates to the field of information processing technology, specifically to a human-computer hierarchical collaborative decision-making and contribution determination method for multidisciplinary education and scientific research. Background Technology

[0002] With the widespread application of generative AI and large-scale models in university teaching and research, human-machine collaboration in completing research projects, curriculum development, teaching reform plans, and industry-education integration projects has become commonplace. Whether it's building engineering training systems, refining business teaching cases, designing humanities curriculum maps, or tackling teaching and research challenges in the wine industry, teachers and researchers use AI to complete tasks such as literature screening, data calculation, and generating initial drafts. The final teaching and research results integrate the researchers' professional judgment, innovative thinking in the discipline, and the data processing and content generation capabilities of artificial intelligence.

[0003] From the perspective of actual management needs in universities, current research performance evaluation, professional title assessment, and project evaluation all require an objective distinction between the innovative value of humans and AI. The confirmation of copyright and intellectual property rights for achievements such as teaching reform papers, teaching research patents, and curriculum resources must also clearly define the proportion of human and machine contributions. Meanwhile, the Ministry of Education has put forward clear requirements for academic integrity and research ethics review, and complete and traceable records of human-machine collaboration are now a mandatory compliance condition. Against this backdrop, a standardized and implementable human-machine collaborative decision-making and contribution confirmation process that is adaptable to multidisciplinary teaching and research scenarios is a real and urgent need for the standardized management of scientific research in universities.

[0004] However, the various research assistance tools and general human-computer collaboration solutions currently available on the market have many unavoidable shortcomings in practical use and cannot meet the specific needs of educational and research scenarios:

[0005] First, the rigid human-machine division of labor model suffers from poor adaptability. Most existing tools fix the responsibilities of humans and machines at the beginning of a project, without any adjustments throughout the process, representing a coarse-grained task-level allocation. However, education and scientific research are inherently step-by-step and highly time-dependent processes. From needs assessment and framework building to indicator assignment and solution optimization, the requirements for human professional judgment and machine computing power assistance vary significantly at different stages. In addition, the evaluation logic and data formats of engineering, business, humanities, and specialized disciplines differ significantly. A fixed division of labor either wastes manpower on simple, repetitive steps or leads to excessive reliance on AI to produce low-quality content in highly specialized areas, greatly reducing overall research efficiency. There is a lack of a hierarchical collaborative approach that can dynamically adjust the human-machine decision-making weights according to the status of each individual task.

[0006] Second, the current methods for evaluating human and machine contributions are one-sided and prone to distorting the results. Neither of the two mainstream methods is suitable for establishing ownership in teaching and research: one only considers superficial data such as computational power consumption and text length, completely ignoring the professional cognitive labor researchers invest in correcting AI errors and reconstructing the research framework; the other relies solely on differences in text characters to determine contribution, a typical "only looking at the final result" approach that fails to consider common real-world scenarios: if the initial AI draft has many logical or professional knowledge flaws, researchers need to spend a significant amount of time making comprehensive adjustments, but simple textual differences cannot reflect this additional cost of innovation. Furthermore, existing solutions do not consider the constraints of flawed preliminary steps on the current research, making it difficult to fairly assess the human value of "repairing innovation based on inferior previous results." Ultimately, performance allocation and authorship can only rely on the subjective judgment of administrators, easily leading to conflicts among researchers.

[0007] Third, the collaborative process lacks a reliable evidence storage link, resulting in shortcomings in compliant evidence collection. Most existing platforms only retain the final product; the basis for each step of the human-machine division of labor, the original AI output, the manual modification trajectory, and the intermediate data for contribution calculation are not fully archived. Furthermore, the centralized storage of data poses a risk of tampering and deletion. In the event of disputes over intellectual property ownership, academic integrity, or performance distribution, the lack of complete and reliable original records as evidence not only fails to meet the requirements for traceability and supervision of scientific research ethics but also poses legal risks to the commercialization of research results and the protection of intellectual property rights.

[0008] Fourth, the general solutions do not align with the specific characteristics of education and research, resulting in poor implementation. Existing human-machine weighting algorithms are mostly designed for industrial simulation and general document generation, lacking specific calculation rules for the three types of data unique to teaching and research scenarios: course ranking sequences, class hour weight indicators, and course prerequisite maps. Furthermore, there is no customizable parameter system for different disciplines. Directly applying these algorithms leads to significant deviations in contribution calculations, making it difficult to adapt to the multidisciplinary and industry-education integration-oriented teaching and research work of higher vocational colleges.

[0009] Continuing to use existing technological solutions will bring multiple management risks: First, the lack of objective and quantifiable benchmarks for human-machine contributions will increase disputes over performance and the distribution of results, thus increasing the burden on scientific research management in universities. Second, the lack of standardized basis for confirming the ownership of research patents, course resources, and other achievements will easily lead to civil ownership disputes in subsequent achievement transformation. Third, the entire process of human-machine interaction cannot be traced, making it difficult to verify ethical issues such as over-reliance on AI and weakening of human innovation, which does not meet the requirements of academic supervision. Fourth, the static division of labor model results in a double waste of human resources and computing power, which is not conducive to the implementation of a new paradigm of education and scientific research driven by artificial intelligence.

[0010] Therefore, the industry urgently needs a set of processing methods for multidisciplinary education and scientific research, which can realize dynamic hierarchical human-machine collaborative decision-making by taking independent sub-steps as units, and combine multiple dimensions such as modification range, cognitive load, and time sequence dependence to hierarchically calculate human-machine innovation contributions. At the same time, it should be equipped with a full-process encrypted evidence storage and smart contract automatic rights confirmation mechanism to make up for the various shortcomings of the existing rigid division of labor, distorted evaluation, difficulty in traceability, and weak discipline adaptation, reduce the legal risks of scientific research management and intellectual property rights in universities, and support the standardization and intelligent governance of education and scientific research in the AI ​​environment. Summary of the Invention

[0011] The purpose of this invention is to address the shortcomings of existing technologies, such as rigid allocation of rights, subjective and one-sided quantification of contributions, untraceable decision-making processes, and single evaluation dimensions. This invention provides a human-computer hierarchical collaborative decision-making and contribution confirmation method for multidisciplinary education and scientific research.

[0012] To achieve the aforementioned objectives, the technical solution adopted in this invention is: a multidisciplinary educational research method for human-computer hierarchical collaborative decision-making and contribution determination, comprising four progressive core stages:

[0013] Phase 1: Standardized Decomposition of Multidisciplinary Teaching and Research Tasks. Identify the subject type and task category of the input teaching and research task. Based on a pre-set subject task ontology library, decompose the complete teaching and research task into an ordered sequence of independent decision-making steps. Label the basic attribute parameters of each step, including the inherent difficulty of the step, prerequisite dependencies, and subject-specific weight configuration.

[0014] Phase 2: State-Aware Human-Machine Layered Collaborative Decision-Making and Execution Phase. For each independent decision-making step, the overall complexity of the step is calculated in real time, taking into account the inherent difficulty of the step, the execution state of the context, and historical interaction data. The human-machine collaboration trust level is updated synchronously and iteratively. Human-machine decision weights are dynamically allocated based on layered threshold judgment logic, distinguishing between three modes: machine-independent execution, human-independent execution, and human-machine hybrid collaborative execution. After the step is completed, a step quality score is calculated based on the output effect, and the quality score is fed back to update the human-machine trust level for subsequent step permission allocation.

[0015] Phase 3: Multi-dimensional Human-Machine Innovation Contribution Layered Quantification Phase. For decision-making steps executed by human and machine hybrid processes, the initial output content of the machine, the final revised output content of the human, and the original data of the human-machine interaction process are collected; the amount of human modification and adjustment is calculated based on the differences in teaching and research data structure types; the cognitive friction coefficient is constructed by introducing the human interaction time to complete the first layer of contribution correction; the decision path dependency degree is constructed by combining the dependencies of the preceding steps to complete the second layer of temporal constraint correction, and the final contribution ratio of human and machine in a single step is obtained; the contributions of all steps are weighted and aggregated by step importance and step execution quality to obtain the total human-machine contribution ratio of the entire teaching and research task.

[0016] Phase 4: Full-process data encryption, evidence preservation, and automatic rights allocation. Complete interactive metadata from all decision-making steps is collected to form standardized decision status records, which are then encrypted using chain hashing to ensure immutable evidence preservation. After the entire task process is completed, the aggregated human-machine contribution ratio is input into a smart contract to automatically allocate the intellectual property rights and research performance benefits of teaching and research achievements.

[0017] Detailed implementation logic for each stage:

[0018] 1. Detailed Implementation of the Standardization and Decomposition Stage for Multidisciplinary Teaching and Research Tasks

[0019] Construct a multidisciplinary education and research task ontology covering engineering, business, and humanities. The ontology stores standardized step templates corresponding to four types of tasks under each discipline: professional development, curriculum development, teaching research, and scientific research projects.

[0020] Input any teaching and research task T, the system matches the ontology database to complete subject classification and task type identification, and outputs an ordered sequence of decision steps. For each step Calibrate the initial difficulty coefficient Labeling rules: Statistically calculate the historical average execution time and frequency of manual intervention for similar steps, and normalize and map them to the 0-1 interval;

[0021] Synchronously record step metadata: unique step identifier, input and output specifications, set of directly dependent preceding steps, parallel execution identifier, and subject-specific weight parameters (indicator importance weight, step type weight).

[0022] 2. Detailed Algorithm for State-Aware Human-Machine Layered Collaborative Decision-Making and Execution Phase

[0023] (1) Real-time comprehensive complexity calculation of the steps

[0024]

[0025] The steps inherently present a challenge;

[0026] The context complexity is derived from the average quality of the preceding steps and is equal to 1 minus the average quality score of the preceding steps. It takes the value of 0 when there are no preceding steps.

[0027] The uncertainty in the execution of a step is equal to the standard deviation of the quality of similar steps in the past.

[0028] , , To assign weights for subject-specific regressions, the sum is 1, and the default configuration is... It supports personalized calibration of least squares regression by discipline.

[0029] (2) Iterative update model of human-machine trust

[0030]

[0031] Initial trust level =0.5, output interval constraint [0.1,0.9];

[0032] For the current k-th (i.e. - After executing the steps and updating the quality score, the new human-machine collaboration trust level is obtained iteratively; after the calculation is completed, it is forcibly truncated to a fixed interval [0.1, 0.9] to prevent the trust level from becoming extreme and causing the human-machine allocation to become completely fixed.

[0033] The historical trust value is the result of the iteration of the previous decision step immediately preceding the current k-step; when k=1 (the first step of the task, with no preceding execution steps), the initial baseline trust value is uniformly substituted. =0.5.

[0034] The output quality deviation of the machine in step k is given. ,in Perform a quality score for the current step. The average quality score for similar historical steps;

[0035] This is the sum of long-term interactive experiences with exponential forgetting decay:

[0036] Forgetting coefficient =0.1;

[0037] , , These are the historical trust retention weight coefficient, the step quality deviation correction weight coefficient, and the long-term interaction experience decay weight coefficient, respectively. All of them can be recalibrated based on the corresponding subject's history teaching and research samples through least squares regression. If there is no custom configuration, the default value is used, which is 0.5 / 0.3 / 0.2.

[0038] Note: If the current step has a quality score If the iterative solution is not yet complete and the problem is in a pending state, then the current trust update will be paused until further notice. Iterative calculations are performed only after convergence.

[0039] (3) Hierarchical threshold dynamic permission allocation logic

[0040]

[0041] The weights for machine decision-making are forcibly truncated to the interval [0,1]. The closer the value is to 0, the more machine-dominated it is; the closer it is to 1, the more human-dominated it is. If the calculated value is less than 0, it is set to 0 (machine executes completely independently); if the calculated value is greater than 1, it is set to 1 (human executes completely independently). , The threshold for complexity stratification is set to a default value. =0.3, =0.7; , The default setting is the trust level tier threshold. =0.7, =0.3, all thresholds support custom configuration by subject.

[0042] (4) Step quality score calculation rules

[0043] Based on human-machine decision weights Three execution scenarios were identified, and corresponding raw interaction data were collected for subsequent iterations of human-machine trust and quantitative calculations of human-machine contributions.

[0044] when =0 (Machine Independent Execution): The complete step plan is generated by AI, and the AI ​​output, model inference confidence, and data support features are recorded simultaneously.

[0045] when =1 (Independent Human Execution): The decision-making and judgment are completed independently by researchers, collecting human judgment logic and finalizing the plan;

[0046] When 0 < <1 (Human-Machine Hybrid Execution): AI first outputs initial content Researchers iteratively revised the initial draft to generate the final revised version. .

[0047] For any step, the quality of step execution is denoted as . Numerical constraint interval ∈[0,1], solve in two ways:

[0048] ① The steps have objective verification indicators: obtained by reverse conversion using normalized error. ;

[0049] ② No objective quantitative verification indicators for the steps: calculations are performed by reverse inference based on the quality results of all subsequent dependent steps.

[0050] If a step has a circular dependency with a post-evaluation step, an iterative algorithm should be used to solve the problem. The algorithm is set with a convergence threshold ε=0.001 and a maximum number of iterations of 20. During the iteration process, the calculation result of each round is limited to the interval [0,1]. If the convergence condition is not met even after the maximum number of iterations is reached, the quality of this step is marked as requiring manual review and calibration.

[0051] 3. Detailed Algorithm for Hierarchical Quantification of Multi-Dimensional Human-Machine Innovation Contributions

[0052] when =0 (Machine-independent execution): Single-step human final contribution =0, machine's final contribution =1;

[0053] when =1 (Executed independently by humans): Final contribution of a single human step =1, Machine's final contribution =0;

[0054] Only 0 < When <1, the human-machine hybrid execution steps are performed, executing a two-layer correction quantization process.

[0055] Specifically

[0056] A standardized decision-making state vector is constructed to store the full-process interactive data, quantifying the amount of human modification, cognitive labor cost, and the impact of time constraints. The final contribution ratio of humans and machines is obtained through two corrections.

[0057] (1) Decision state vector Standardized definition:

[0058]

[0059]

[0060]

[0061] Complete storage steps and full-dimensional metadata serve as original evidence for rights confirmation and auditing.

[0062] (2) Calculation of human adjustment amount in multi-layer structure

[0063] For the three standardized data structures in teaching and research, a dedicated difference measurement algorithm is matched for each, and the difference values ​​of each structure are weighted and fused to obtain the unified human adjustment amount for a single step. :

[0064]

[0065] ① Sequence-type data (e.g., course order, review list): Calculate the structural difference value using Kendall's tau inversion distance and the LCS longest common subsequence combination distance. ;

[0066] ② Key-value data (e.g., lesson weights, indicator scores): Difference values ​​are calculated using subject-weighted Euclidean distance. Weighting of indicators for different disciplines ;

[0067] ③ Directed acyclic graph data (e.g., course prerequisite graphs, project dependencies): use normalized graph edit distance to calculate the difference value. The extent of structural changes is represented by the proportion of symmetrical differences between nodes and newly added / missing edges;

[0068] Always , , The weights of the built-in data structure for each step sum to 1. These weights are automatically matched and assigned during the S1 task standardization and decomposition phase. When the human-machine output structure is completely heterogeneous... =1 indicates that the human has completely reconstructed the output. If only a local data structure mismatch occurs, only the heterogeneous part is assigned a value, and the adjustment range for the remaining structure is calculated normally.

[0069] (3) Coefficient of cognitive friction Used to quantify artificially modified cognitive load

[0070]

[0071] The actual interaction time for manually revising the AI ​​draft;

[0072] This serves as the baseline value for the lower quartile of the historical modification time for similar steps.

[0073] A score greater than 1 indicates a high cognitive load and significant costs for manual correction. <1 indicates high AI output quality, requiring only minor adjustments by human intervention;

[0074] (4) Decision path dependence Used to characterize the defect constraints of the preceding steps

[0075]

[0076] : The set of all preceding dependent steps of the current step; This represents the total number of preliminary steps.

[0077] Preliminary steps For the current step The constraint weights are pre-configured in the task template;

[0078] Preliminary steps Execution quality score Characterizing this preliminary step The extent of quality defects;

[0079] The path dependence value is obtained by combining the constraint weights of all preceding steps and the magnitude of defects. The larger the value, the stronger the impact of the preceding defects that need to be remedied, and the greater the contribution of human innovation needs to be.

[0080] (5) Contribution-based two-layer correction algorithm

[0081] First layer: Basic contributions based on cognitive friction correction

[0082] Basic contributions:

[0083] Set tolerance threshold 0 = 0.15, high friction amplification factor 0 = 0.3, low frictional attenuation coefficient 0 = 0.2;

[0084]

[0085] The numerical range is truncated to [0,1] to prevent the contribution ratio from exceeding a reasonable range.

[0086] Second layer: Secondary correction based on path dependency time series

[0087] Pre-cumulative error The formula is used to uniformly quantify the overall quality defect magnitude of all preceding steps:

[0088]

[0089] In the formula This is the set of all preceding dependent steps for the current step. This represents the total number of preceding steps; Preliminary steps Perform quality scoring. This preliminary step The extent of quality defects.

[0090] Humanity's final single-step contribution:

[0091]

[0092] Calculate the product term The upper limit is truncated to 0.3 to prevent the value from exceeding 1 after adjustments for human contributions;

[0093] Machine's final single-step contribution:

[0094] .

[0095] After the calculation is completed, finally... , Cut off to the interval [0,1].

[0096] (6) Aggregation of total contribution across all tasks

[0097]

[0098]

[0099] The importance weights for each step are summed to 1; due to aggregation... , The sum of the two weighted original contribution values ​​is not necessarily equal to 1. Normalization is required before outputting the weighted allocation ratio. The final weighted ratio is:

[0100] ÷( + )

[0101] ÷( + );

[0102] 4. Detailed Implementation of Encrypted Evidence Preservation and Smart Contract Rights Confirmation

[0103] (1) Step-by-step hash chain evidence storage:

[0104] For each step, a standardized Decision State Vector (DSV) is constructed to store the step identifier, executing entity, human-machine output, all intermediate calculation parameters, quality score, and contribution value; a unique summary is generated.

[0105] Serial SHA-256 hash chaining is used for evidence storage: (in the formula) (Represents string concatenation operation), with the genesis hash as the starting point of the chain. All step hashes are concatenated onto the chain, and the data cannot be tampered with.

[0106] Once the task is completed, the total contribution ratio will be written into a preset smart contract, automatically mapping the research points and intellectual property ownership allocation ratios, and outputting the allocation results through the university's research management system.

[0107] The beneficial effects of this invention are mainly reflected in:

[0108] 1. Step-level dynamic hierarchical permission allocation adapts to the differentiated teaching and research needs of multiple disciplines in higher vocational colleges; it breaks the traditional task-level static human-machine responsibility division model, takes independent decision-making steps as the smallest unit, and dynamically adjusts the AI ​​decision weight by integrating real-time task complexity and human-machine historical interaction trust; it configures exclusive calculation parameters for different disciplines, solves the problem of human-machine adaptation imbalance in multiple types of teaching and research tasks, and significantly improves the efficiency of collaborative decision-making.

[0109] 2. Multi-structure hierarchical editing distance accurately quantifies the contribution of manual modifications, and the evaluation criteria are objective and reproducible; differentiated distance measurement algorithms are designed for three mainstream data types in higher vocational education and research scenarios: sequences, key values, and knowledge graphs, replacing traditional fuzzy subjective estimation; all calculation logic has standardized mathematical formulas, the contribution quantification process is reproducible, and it provides objective and verifiable numerical basis for the intellectual property rights confirmation of human-computer collaborative education and research results.

[0110] 3. Dual innovation quantitative indicators, taking into account both the extent of modification and the value of human cognitive labor, avoiding the bias of solely focusing on results; This invention introduces for the first time the cognitive friction coefficient to quantify the cognitive cost borne by humans in correcting inferior AI output, while using decision path dependence to characterize the decision constraints brought about by defects in the preceding steps; The dual-layer correction mechanism can reasonably amplify the human contribution of the "deep reconstruction scheme based on low-quality AI draft", objectively distinguish between simple fine-tuning and deep innovative labor, and make the evaluation results more fair and in line with the actual teaching and research.

[0111] 4. The entire process is stored on a hash chain blockchain for evidence preservation, and a complete audit chain resolves ownership disputes; the system retains all original data of each human-computer interaction, intermediate parameters of contribution calculation, and permission switching records, and achieves data immutability by connecting the SHA-256 hash chain; when disputes arise regarding the ownership of intellectual property rights of teaching reform papers, teaching and research patents, and curriculum resources, the entire decision-making process can be fully traced, providing credible audit evidence. Attached Figure Description

[0112] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0113] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. It should be noted that these embodiments are only used to illustrate the present invention and should not be regarded as limiting the scope of the present invention.

[0114] Example 1: Teaching and Research Task of "Revision of Talent Training Program for Computer Application Technology Major in Higher Vocational Colleges"

[0115] This embodiment takes the teaching and research task of "Revision of the Talent Training Program for Computer Application Technology" in the School of Information Engineering of a higher vocational college as an example to illustrate the complete implementation process of the method of the present invention.

[0116] (I) S1 stage: Standardized decomposition of multi-disciplinary teaching and research tasks

[0117] Step S1.1: Subject type and task category identification

[0118] The system receives the input task description document, "Revision of the Talent Training Program for the 2026 Class of Computer Application Technology Major". The system calls the natural language processing module to extract keywords and perform semantic classification on the task description, identifying the discipline type of the task as "Engineering" (Computer Science) and the task category as "Professional Development Decision".

[0119] Step S1.2: Retrieve the task ontology library and match it with the step template.

[0120] The system retrieves the corresponding standardized step template from a pre-built multidisciplinary teaching and research task ontology based on the discipline type "Engineering" and the task category "Professional Development Decision". The standardized step template for the Engineering Professional Development Decision task in the ontology is predefined as the following eight steps:

[0121]

[0122] The inherent difficulty of each step Based on statistical analysis and calibration of historical execution data of similar steps in the system database, the average execution time of similar steps in history was statistically determined. Frequency of human intervention ,according to =0.5·norm( )+0.5·norm( After calculation, it is normalized to the interval [0,1], and norm(·) is the Min-Max normalization function.

[0123] The inherent difficulty of step S1 =0.15 indicates that this step is a data-intensive task with strong machine processing capabilities and low need for human intervention; the inherent difficulty of step S7. =0.60 indicates that this step involves multi-dimensional comprehensive judgment and has a high degree of reliance on human professional experience.

[0124] Subject-specific weights for each step (Step importance weights) are preset by the ontology library based on the engineering professional construction scenario: =0.08, =0.07, =0.12, =0.15, =0.20, =0.10, =0.18, =0.10, satisfying Σ =1.

[0125] The prerequisite dependencies, Deps(sᵢ), for each step are recorded as follows: =0 (no prerequisite steps depend on it, and the cumulative error of the prerequisite steps is also 0). =0 (no prerequisite steps depend on it, and the cumulative error of the prerequisite steps is also 0). =[s1,s2], =[s3], =[s4], =[s5], =[s5,s6], =[s7].

[0126] (II) S2 Phase: State-Aware Human-Machine Hierarchical Collaborative Decision-Making and Execution

[0127] Step S2.1: Execute the steps —Industry talent demand data analysis

[0128] (1) Real-time complexity of calculation steps

[0129] Initialize human-machine trust level =0.5. Steps There are no prerequisite steps, therefore =0. Query the system database to find the standard deviation of the execution quality of similar "industry talent demand data analysis" steps in the past. =0.08. Weight parameters calibrated using least squares regression in an engineering setting. Substitute into the formula:

[0130] =0.4×0.15+0.35×0+0.25×0.08=0.060+0+0.020=0.080

[0131] (2) Calculate the permission allocation coefficient

[0132] current =0.5, =0.080< =0.3, but T(0)=0.5 does not satisfy the condition. > The condition = 0.7 is met, therefore the machine-independent execution branch is not triggered. Meanwhile... =0.080 does not meet the requirement. =0.7 condition, =0.5 does not satisfy the condition. < The condition = 0.3 is met, therefore the human-machine hybrid execution branch is entered:

[0133] = ·( - ) / ( - = 0.5 × (0.080 - 0.3) / (0.7 - 0.3) = 0.5 × (-0.22) / 0.4 = -0.275

[0134] Since the calculation result is negative, according to the truncation rule... =0, meaning step It is executed independently by the machine.

[0135] (3) Execution steps

[0136] The system utilizes an AI data analysis engine to automatically collect IT industry recruitment data, job skill requirements data, and salary distribution data from the past three years, generating an "Industry Talent Demand Data Analysis Report." (Machine output) ( It includes structured data such as a list of the top 20 job openings, frequency distribution of skill keywords, and salary range statistics.

[0137] (4) Calculation of the quality score of the steps

[0138] step An objective verification standard exists—data source completeness verification. The system checks that the data sources collected by AI cover mainstream recruitment platforms (BOSS Zhipin, 51job, Lagou.com) and industry association reports, with a data completeness of 95%. The normalization error is 0.05, therefore:

[0139] =1-0.05=0.95

[0140] (5) Update human-machine trust level

[0141] step Executed independently by the machine. Query the system database for the historical average quality score of similar steps. =0.82, then (1) = 0.95 - 0.82 = 0.13.

[0142] Calculate E(1): Since k=1, there are no preceding steps, so E(1)=0.

[0143] Substitute into the trust update formula (calibration parameters for engineering scenarios) =0.5, =0.3, =0.2):

[0144] T(1)=0.5×T(0)+0.3×ΔP(1)+0.2×E(1)=0.5×0.5+0.3×0.13+0=0.250+0.039=0.289

[0145] Cut off to the interval [0.1, 0.9], T(1) = 0.289.

[0146] Step S2.2: Execute the steps —Graduate Employment Quality Survey

[0147] (1) Real-time complexity of calculation steps

[0148] =0.4×0.20+0.35×0+0.25×0.10=0.080+0+0.025=0.105

[0149] (Standard deviation of quality for similar steps in history) =0.10)

[0150] (2) Calculate the permission allocation coefficient

[0151] T(1) = 0.289, =0.105< =0.3, but T(1)=0.289 does not satisfy the condition. > The condition is 0.7. Enter the human-machine hybrid execution branch:

[0152] =0.289×(0.105-0.3) / (0.7-0.3)=-0.141

[0153] After cutting =0, the machine executes independently.

[0154] (3) Execution steps And calculate the mass

[0155] The system uses an AI engine to extract employment data from the school's employment system for the past three years, generating an employment quality analysis report. After data integrity verification, a quality score is awarded. =0.88.

[0156] (4) Update human-machine trust level

[0157] ΔP(2)=Q(2)- =0.88-0.82=0.06.

[0158] E(2)=exp(-0.1×(2-1-0))·Q(1)=exp(-0.1)×0.95=0.9048×0.95=0.8596

[0159] T(2)=0.5×T(1)+0.3×ΔP(2)+0.2×E(2)=0.5×0.289+0.3×0.06+0.2×0.8596=0.1445+0.018+0.1719=0.3344

[0160] Step S2.3: Execute the steps — Diagnostic evaluation of the existing curriculum system

[0161] (1) Real-time complexity of calculation steps

[0162] step Dependency Pre-steps and Context complexity =1-avg(Q( ),Q( =1-(0.95+0.88) / 2=1-0.915=0.085.

[0163] Standard deviation of quality for similar steps in history =0.15.

[0164] =0.4×0.45+0.35×0.085+0.25×0.15=0.2473

[0165] (2) Calculate the permission allocation coefficient

[0166] T(2) = 0.3344, =0.2473< =0.3, but T(2)=0.3344 does not satisfy T> =0.7. Entering the human-machine hybrid execution branch:

[0167] =0.3344×(0.2473-0.3) / (0.7-0.3)==-0.0441

[0168] After cutting =0, the machine executes independently.

[0169] (3) Execution steps And calculate the mass

[0170] The system calls upon the AI-powered teaching evaluation engine, based on and The output data is used to diagnose and evaluate the existing curriculum system, providing scores on the matching degree between each course and industry needs, course overlap analysis, and identification of missing courses. The quality score is verified by academic affairs experts through random checks. =0.82.

[0171] (4) Update human-machine trust level

[0172] ΔP(3) = 0.82 - 0.82 = 0.

[0173] E(3)=exp(-0.1×1)×Q(2)+exp(-0.1×2)×Q(1)=0.9048×0.88+0.8187×0.95=0.7962+0.7778=1.5740

[0174] T(3)=0.5×T(2)+0.3×0+0.2×1.5740=0.1672+0+0.3148=0.4820

[0175] Step S2.4: Execute the steps — Job Competency Standards Development

[0176] (1) Real-time complexity of calculation steps

[0177] =1-Q(s3)=1-0.82=0.18. u(s4)=0.12.

[0178] =0.4×0.50+0.35×0.18+0.25×0.12=0.293

[0179] (2) Calculate the permission allocation coefficient

[0180] T(3) = 0.4820, R(s4) = 0.293 < =0.3, but T(3)=0.4820 does not satisfy T> =0.7. Entering the human-machine hybrid execution branch:

[0181] =0.4820×(0.293-0.3) / (0.7-0.3)=-0.0084

[0182] After cutting =0, the machine executes independently.

[0183] (3) Execution steps And calculate the mass

[0184] The system is based on The curriculum system diagnostic results automatically generated a draft of the Computer Application Technology professional job competency standards, covering a competency matrix for eight core positions. After review by the professional leader, the quality score was Q( =0.78.

[0185] (4) Update human-machine trust level

[0186] ΔP(4) = 0.78 - 0.82 = -0.04.

[0187] E(4)=exp(-0.1×1)×Q(3)+exp(-0.1×2)×Q(2)+exp(-0.1×3)×Q(1)=0.7419+0.7205+0.7038=2.1662

[0188] T(4)=0.5×T(3)+0.3×(-0.04)+0.2×2.1662=0.6622

[0189] Step S2.5: Execute the steps —Course Adjustment Plan Generation

[0190] (1) Real-time complexity of calculation steps

[0191] =1-Q(s4)=1-0.78=0.22. u( =0.18.

[0192] =0.4×0.55+0.35×0.22+0.25×0.18=0.342

[0193] (2) Calculate the permission allocation coefficient

[0194] T(4) = 0.6622, =0.342. Because =0.3< =0.342< =0.7, and T(4)=0.6622 is between =0.3 and When the value is between 0.7, the system enters the human-machine hybrid execution branch:

[0195] =T(4)·(R( )- ) / ( - = 0.6622 × (0.342 - 0.3) / (0.7 - 0.3) = 0.0695

[0196] =0.0695 indicates that the machine decision-making weight is about 7% and the human decision-making weight is about 93% in this step, which belongs to the human-machine hybrid execution mode, with humans playing a dominant role.

[0197] (3) Execution steps

[0198] The system's AI engine is based on Based on the job competency standards, three draft course adjustment plans were automatically generated. ( ):

[0199] Option A: Add a course on "Applications of Artificial Intelligence" and adjust the class hours for five courses, including Data Structures and Database Principles.

[0200] Option B: Reconstruct the curriculum modules into a three-tier system: "Basic Competency Layer - Core Competency Layer - Comprehensive Application Layer".

[0201] Option C: Reorganize 12 core courses with project-based learning as the main approach.

[0202] The expert leader (human decision-maker) reviewed three plans in the system's interactive interface, selected Plan B as the basic framework, and made the following adjustments:

[0203] The "Comprehensive Application Layer" is divided into two levels: the "Comprehensive Application Layer" and the "Innovation Expansion Layer".

[0204] The semester order of 5 courses has been adjusted.

[0205] The credit hour allocation for three courses has been revised ("Web Front-End Development" has been increased from 48 credit hours to 64 credit hours, and "Mobile Application Development" has been increased from 64 credit hours to 48 credit hours).

[0206] Humanity's final output This is the revised three-tiered curriculum system plan.

[0207] (4) Calculate the execution quality Q( )

[0208] step There is no direct objective verification standard; it requires subsequent dependency steps. and The quality is reverse-evaluated. Since subsequent steps have not yet been performed, the system temporarily records Q(…). (To be determined) and After execution, backtracking calculations are performed.

[0209] (5) Trust level will not be updated for now (pending Q( (Updated after confirmation), meaning because... The step quality score is in an iterative solution pending state, triggering the trust level pause update rule, pending... , After execution and iterative convergence, Q( Then, the global human-computer collaboration trust level will be updated uniformly.

[0210] Step S2.6: Execute the steps —Evaluation of Teaching Resource Allocation

[0211] =0.4×0.40+0.35×(1-Q( (pending)+0.25×0.10

[0212] Due to Q( (Not yet determined, system usage) The mass is used as an approximation: ≈1-Q( =0.22.

[0213] =0.4×0.40+0.35×0.22+0.25×0.10=0.262

[0214] T(4) = 0.6622, =0.262< =0.3, enter the human-machine hybrid branch:

[0215] =0.6622×(0.262-0.3) / 0.4=-0.0629, after truncation =0, the machine executes independently.

[0216] System call resource evaluation engine, based on The solution automatically assesses teacher matching, lab capacity, equipment requirements, etc., and outputs a resource allocation assessment report. =0.85.

[0217] Step S2.7: Execute the steps — Feasibility study of the plan

[0218] =0.4×0.60+0.35×(1-avg(Q( pending, Q( )))+0.25×0.14 used and Approximate value: c≈1-(0.78+0.85) / 2=0.185.

[0219] =0.24 + 0.06475 + 0.035 = 0.3398

[0220] =0.6622×(0.3398-0.3) / 0.4=0.6622×0.0398 / 0.4=0.0659

[0221] Human-machine hybrid execution =0.0659 (machine weight approximately 6.6%, human weight approximately 93.4%).

[0222] Step S2.8: Execute the steps —Final review and approval of the plan

[0223] =0.4×0.35+0.35×(1-avg(Q( ),Q( )))+0.25×0.10

[0224] Use Q( )=0.85 and Q( The approximate value to be determined is: c≈1-(0.85+0.80) / 2=0.175.

[0225] =0.14 + 0.06125 + 0.025 = 0.2263

[0226] =0.6622×(0.2263-0.3) / 0.4=-0.122, after truncation =0, the machine executes independently.

[0227] The system automatically summarizes the output of the aforementioned steps, generates a complete final revised plan, and confirms it online through the professional construction committee. =0.92.

[0228] Step S2.9: Backtrack to calculate Q( ) and Q( )

[0229] step The direct dependence of subsequent steps is and But Q( Dependence on Q( ) and Q( Iterative solution is used:

[0230] Initial hypothesis Q( )=0.80, Q( =0.80.

[0231] Round 1: Q( )=1-( (1-Q( ))+(1-Q( )) ) / 2 = 1-(0.15+0.20) / 2 = 0.825

[0232] Q( )=1-( (1-Q( ))+(1-Q( )) ) / 2 = 1-(0.175+0.15) / 2 = 0.8375

[0233] Iterate until convergence: Q( )=0.83, Q( =0.84.

[0234] The final quality score for each step is: Q( ) to Q( The values ​​were 0.95, 0.88, 0.82, 0.78, 0.83, 0.85, 0.84, and 0.92, respectively.

[0235] Step S2.10: Final Trust Update

[0236] =0.83-0.82=0.01

[0237] =0.84-0.82=0.02

[0238] =0.92-0.82=0.10

[0239] E(8)≈4.12.

[0240] T(8)≈1.229, cut off to 0.9.

[0241] (III) S3 stage: Multi-dimensional quantification of human-machine innovation contributions

[0242] Step S3.1: Calculation steps Human adjustment amount

[0243] step The output is the curriculum scheme, which includes three types of data structures:

[0244] Sequence data: Semester-ordered list of 12 core courses

[0245] Key-value data: Key-value pairs representing the credit hour allocation for each course.

[0246] Graph structure data: Course prerequisite relation DAG (12 nodes, 15 directed edges)

[0247] (1) Calculate the adjustment amount for sequential data

[0248] Machine output The course order is as follows: [Data Structures, Database Principles, Web Front-end Development, Mobile Application Development, Software Engineering, Operating Systems, Computer Networks, Algorithm Analysis, Project Management, Cloud Computing Fundamentals, Introduction to Artificial Intelligence, Graduation Project]

[0249] Humanity's ultimate The course order is as follows: [Data Structures, Database Principles, Web Front-End Development, Software Engineering, Operating Systems, Computer Networks, Algorithm Analysis, Project Management, Mobile Application Development, Cloud Computing Fundamentals, Introduction to Artificial Intelligence, Graduation Project]

[0250] The two sequences have a length of n=12, and are permutations of elements without repetition. =1, =0. Statistical inversions: The order changed only between the courses "Mobile Application Development" and "Software Engineering," "Operating Systems," "Computer Networks," "Algorithm Analysis," and "Project Management," resulting in 5 inversion pairs. The total number of element pairs is C(12,2)=66.

[0251] =5 / 66=0.0758

[0252] =1×0.0758+0=0.0758

[0253] (2) Calculate the adjustment amount of key-value data

[0254] The system allocates class hours for 12 courses, using class hours (unit: class hours) as the key. In engineering scenarios, the system pre-sets a higher γ value for the key related to "practical class hour percentage". To simplify calculations, it assumes that all keys have equal weight. =1 / 12.

[0255] The machine output and the final human output changed over the course of five courses. The weighted Euclidean distance was calculated (detailed numerical calculations omitted). =0.0835.

[0256] (3) Calculate the amount of data adjustment for the graph structure.

[0257] In the prerequisite relation DAG, the machine output has 12 nodes and 15 edges, while the human output has 12 nodes and 17 edges (2 prerequisite relations were added).

[0258] | \ |=0,| \ |=0,| \ |=2 (edges that machines have but humans don't), | \ |=4 (Human-added edges, of which 2 are brand new and 2 replace machine edges).

[0259] =(0+0+2+4) / (12+12+15+17)=6 / 56=0.1071

[0260] (4) Calculation steps and comprehensive adjustment amount

[0261] step The data structure type weight is =0.5, =0.2, =0.3 (The curriculum system is mainly arranged in sequence).

[0262] =0.5×0.0758+0.2×0.0835+0.3×0.1071=0.0867

[0263] That is, humans in steps The basic modifications and adjustments made to the machine output are approximately 8.67%.

[0264] Step S3.2: Calculate the basic contribution

[0265] ( )= =0.0867

[0266] =1-0.0867=0.9133

[0267] Step S3.3: Calculate the cognitive friction coefficient (CFC). )

[0268] The system records the steps taken by the professional leader. Actual interaction time ( = 47 minutes (from opening the interactive interface to view the three solutions to submitting the final modified solution).

[0269] Querying the system database, the lower quartile of the historical modification time for the "Course Adjustment Plan Generation" step of the same type. ( = 32 minutes (meaning that historically, 75% of similar steps took more than 32 minutes to modify, and 25% took less than 32 minutes).

[0270] CFC ) = 47 / 32 = 1.46875

[0271] CFC The value of 1.47 > 1 + δ = 1.15 indicates a high cognitive friction scenario, suggesting that the three solutions generated by AI deviate significantly from the professional judgment of human experts, requiring humans to expend considerable cognitive effort to correct them.

[0272] Step S3.4: First-level contribution correction (based on cognitive friction)

[0273] ( )= ( )·(1+α·(CFC( )-1))=0.0867×(1+0.3×(1.47-1))=0.0989

[0274] ( =1 - 0.0989 = 0.9011

[0275] Step S3.5: Calculate the decision path dependency degree (DPD). )

[0276] step Prerequisite Dep( )={ The constraint weight of the preceding step on the current step is ω(s4,s5)=0.75 (the job competency standard has a high degree of constraint on the course adjustment plan).

[0277] DPD( )=(1 / |{ }|)·ω( , )·(1-Q( =1 × 0.75 × (1 - 0.78) = 0.165

[0278] Step S3.6: Calculate the preceding cumulative error

[0279] =(1 / |{ }|)·(1-Q( =1 - 0.78 = 0.22

[0280] Step S3.7: Second-layer contribution correction (based on path dependency)

[0281] = ( )·(1+DPD( )· = 0.0989 × (1 + 0.165 × 0.22) = 0.1025

[0282] =1-0.1025=0.8975

[0283] Step S3.8: Calculate the contribution of other mixed execution steps.

[0284] step Also for mixed execution, the same calculation method is used (detailed numerical calculation process omitted), resulting in:

[0285] =0.0873, =0.9127

[0286] Step S3.9: Assigning contribution values ​​for independently executed steps

[0287] For steps executed independently by the machine , , , , , :

[0288] =0, =1

[0289] Step S3.10: Aggregate total contribution across all tasks

[0290]

[0291] =0.83×0.1025×0.20+0.84×0.0873×0.18

[0292] =0.0170 + 0.0132 = 0.0302

[0293]

[0294] (The contribution of each step is 1 when the machine executes it independently.) and (minus human portion)

[0295] Calculate the weighted sum of quality for all steps: 0.95×0.08+0.88×0.07+0.82×0.12+0.78×0.15+0.83×0.20+0.85×0.10+0.84×0.18+0.92×0.10=0.076+0.0616+0.0984+0.117+0.166+0.085+0.1512+0.092=0.8472

[0296] =0.8472-0.0302=0.8170

[0297] The final human-machine contribution ratio, calculated using the formula, is approximately 3.56%:96.44%.

[0298] In the task of revising this talent training program, the total contribution of human experts was approximately 3.56%, while the total contribution of AI systems was approximately 96.44%.

[0299] (iv) S4 phase: Full-process data encryption and storage and automatic rights confirmation and allocation

[0300] Step S4.1: Construct the decision state vector (DSV) for each step.

[0301] In steps For example, the system constructs the following decision state vector:

[0302] DSV( )={ .id:"T2026-CS-PLAN-STEP5",executor( ):"hybrid",timestamp( ):"2026-03-15T14:23:17+08:00",dependencies(s5):["s4"],

[0303] ( {Complete structured data for schemes A, B, and C}

[0304] ( ):{Complete structured data of the modified three-tier curriculum system scheme},

[0305] ( ):0.0867,

[0306] CFC ):1.46875,

[0307] DPD( ):0.165,

[0308] 0.0867,

[0309] 0.0989,

[0310] 0.1025,

[0311] 0.8975,

[0312] ( ):{Final Three-Tier Curriculum System Scheme},

[0313] Q( ):0.83,

[0314] λ( ):0.0695}

[0315] Step S4.2: Chained hash encryption for evidence storage

[0316] Setting the Genesis Hash

[0317] Hash( )="0000000000000000000000000000000000000000000000000000000000000000000000000" .

[0318] Hash(s1) = SHA-256(DSV(s1) || Hash(s0))

[0319] Hash(s2) = SHA-256(DSV(s2) || Hash(s1)) ...

[0321] Hash(s8) = SHA-256(DSV(s8) || Hash(s7))

[0322] The hash values ​​from each step are chained together to form a complete chain of evidence. Any tampering with data in any DSV will cause changes to all subsequent hash values, thus triggering system detection.

[0323] Step S4.3: Automatic allocation of rights and interests in smart contracts

[0324] After the entire task process is completed, the system will aggregate the total contribution ratio of humans and machines (…). : =0.0302:0.8170) is written into the preset smart contract.

[0325] According to the school's research performance management regulations, this achievement receives 100 research performance points. Smart contract automatic execution allocation:

[0326] Human contributors (Professional Construction Committee) receive: 100 × 0.0302 / (0.0302+0.8170) = 100 × 0.0356 = 3.56 points

[0327] The AI ​​system (the teaching and research AI platform purchased by the school) received: 100 × 0.8170 / (0.0302+0.8170) = 100 × 0.9644 = 96.44 points (this part is under the unified management of the school's scientific research information department).

[0328] The allocation results are automatically pushed to the school's scientific research management system as the basis for annual scientific research performance evaluation.

[0329] Example 2: Teaching and Research Task of "Development of Core Curriculum Standards for E-commerce Major in Higher Vocational Colleges"

[0330] This embodiment takes the task of "Developing the Course Standards for the E-commerce Major's 'Network Marketing' Course" in the Business School of a higher vocational college as an example to further illustrate the differentiated implementation of the method of the present invention in different disciplinary scenarios.

[0331] (a) Differences in subject-specific scenarios

[0332] Unlike the engineering major "Computer Application Technology" in Example 1, this example is the business major "E-commerce". In stage S1, the system identifies the discipline type as "Business" and automatically retrieves the specific parameter configurations developed for the business curriculum standards from the ontology database.

[0333] In key-value data, the γ values ​​of keys such as "case teaching weight", "practical training simulation ratio", and "enterprise cooperation project weight" are preset to relatively high values;

[0334] Trust update model , , The weighting parameters were independently calibrated using historical business school samples;

[0335] Each step , , The weights are adjusted based on the characteristics of the business curriculum standards (business curriculum standards contain more textual descriptive content, while sequential data has relatively lower weights).

[0336] (II) Examples of Key Steps

[0337] In the core step of curriculum standard development, "Design of Curriculum Content and Teaching Methods":

[0338] Initial machine output: Based on the national teaching standards database and industry talent demand data, the AI ​​generates a course content framework for "Internet Marketing", which includes 8 teaching units, teaching objectives for each unit, and suggested class hours.

[0339] Human modification process: Two professional teachers collaborated to modify the system interface, merging the two independent units of "Search Engine Marketing" and "Social Media Marketing" suggested by AI into a "New Media Marketing Integration" unit, adding a "Live Streaming E-commerce Practice" unit, adjusting the teaching method descriptions of 5 units, and changing the AI-recommended "lecture + case analysis" model to a "project-driven + flipped classroom" model.

[0340] H_adj calculation: Adjustment amount for sequence data (ordering of teaching units) =0.12; Adjustment amount for key-value data (study hour allocation) =0.18; Adjustment amount for graph structure data (knowledge association graph) =0.22. Overall =0.15.

[0341] CFC calculation: =65 minutes (Quarters of quartiles for similar steps in business studies) = 42 minutes, CFC = 1.55. High cognitive friction amplifies and corrects for human contributions.

[0342] DPD Calculation: The quality of the preliminary step "Learning Situation Analysis" is Q=0.72, the constraint weight is ω=0.60, and DPD=0.60×(1-0.72)=0.168. Preliminary Cumulative Error =0.28, human contribution is amplified by a second correction.

[0343] Final contribution: After two layers of correction, the final human contribution for this step. ≈0.18 (18%), reflecting a higher degree of involvement of human professional judgment in the development of business curriculum standards.

[0344] (III) Subject Adaptation Effect

[0345] Because the parameters of the trust update model in business scenarios differ from those in engineering scenarios (business scenarios place more emphasis on the weight of recent interaction experience E(k), the system tends to assign steps requiring high levels of professional judgment to human-led execution in subsequent steps, making the overall human-machine division of labor more in line with the practical characteristics of business teaching and research.

[0346] Implementation conditions and system deployment instructions

[0347] The implementation of the method of this invention depends on the following hardware and software environment:

[0348] (a) Hardware environment

[0349] The system is deployed on the school's private cloud server or hybrid cloud architecture, with a configuration of no less than: 16 CPU cores, 64GB memory, 2TB storage space (SSD), and a GPU accelerator card.

[0350] (ii) Software Environment

[0351] Operating System: Linux (Ubuntu 20.04 LTS or above); Database: PostgreSQL 13+ (for storing structured decision data) + IPFS (for storing unstructured interactive logs); Blockchain Network: Hyperledger Fabric or Ethereum private chain; AI Engine: Inference service supporting mainstream deep learning frameworks (TensorFlow / PyTorch).

[0352] (III) Data Preparation

[0353] Before deployment, the following must be completed: Construction and annotation of a multidisciplinary teaching and research task ontology (covering at least three disciplines: engineering, business, and humanities, with at least four task templates for each discipline); regression calibration of trust update model parameters for each discipline scenario (at least 100 historical decision-making step samples for each discipline); and historical baseline duration for each step type. Statistical calculations (no less than 30 historical samples for each step).

[0354] (iv) User terminal

[0355] Teachers can access the system through a web browser or a dedicated client. The interactive interface supports common operations in teaching and research scenarios, such as online document editing, drag-and-drop sorting, and relationship diagram drawing. All operations are automatically recorded as interaction logs by the system.

Claims

1. A human-computer hierarchical collaborative decision-making and contribution determination method for multidisciplinary educational research, characterized in that, Includes the following steps: S1. Standardized decomposition stage of multi-disciplinary teaching and research tasks: Identify the subject type and task category corresponding to the teaching and research task to be processed, retrieve the pre-constructed multi-disciplinary teaching and research task ontology library, decompose the complete teaching and research task into an ordered sequence of independent decision-making steps, and label each decision-making step with basic attribute parameters including inherent difficulty, prerequisite dependencies, and subject-specific weights. S2, State-Aware Human-Machine Layered Collaborative Decision-Making Execution Stage: For each independent decision-making step, the overall complexity of the step is calculated by combining the inherent difficulty of the step, the context execution state, and historical human-machine interaction data, and the human-machine collaboration trust level is iteratively updated; the human-machine decision weights are dynamically allocated based on the layered threshold judgment logic, and three execution modes are divided: machine independent execution, human independent execution, and human-machine hybrid collaborative execution; after the step is executed, the step quality score is calculated based on the output effect, and the step quality score is fed back to update the human-machine collaboration trust level of subsequent steps; S3, Multi-dimensional Human-Machine Innovation Contribution Layered Quantification Stage: If step S2 is executed independently by the machine or by a human, then the corresponding human-machine contribution ratio is directly assigned. For the decision-making step in step S2, which involves human-machine collaborative execution, the initial machine output, the revised human output, and the complete raw data of human-machine interaction are collected. Based on the differences in the teaching and research data structure, the extent of human modification and adjustment to the machine output is quantified. Combined with the human interaction time, a cognitive load index is constructed to complete the first layer of contribution correction. Then, combined with the dependency relationship of the preceding steps, a time-series constraint index is constructed to complete the second layer of contribution correction, thus obtaining the human-machine contribution ratio for a single step. The human-machine contributions of all decision-making steps are weighted and aggregated based on the importance of the steps and the quality of step execution to obtain the total human-machine contribution ratio for the complete teaching and research task. S4. Full-process data encryption and automatic rights allocation stage: Collect complete interactive metadata of all decision-making steps to generate standardized decision status records, and use chain hash encryption to complete tamper-proof evidence storage; after the entire teaching and research task is completed, input the aggregated human-machine total contribution ratio into the smart contract, and automatically allocate teaching and research results according to the contribution ratio.

2. The human-computer hierarchical collaborative decision-making and contribution determination method for multidisciplinary educational research according to claim 1, characterized in that: In step S2, the human-machine collaboration trust level is iteratively updated based on the execution effect of the previous steps, and the trust level value is constrained within a fixed range to avoid extreme permission allocation.

3. The human-computer hierarchical collaborative decision-making and contribution determination method for multidisciplinary educational research according to claim 1, characterized in that: In step S2, the hierarchical threshold determination logic jointly determines the human-machine decision weight based on the step comprehensive complexity threshold and the human-machine collaboration trust threshold.

4. The human-computer hierarchical collaborative decision-making and contribution determination method for multidisciplinary educational research according to claim 1, characterized in that: In step S3, three types of educational and research data are distinguished: sequential, key-value, and graph-structured. A differentiated distance metric is used to calculate the extent of human modification and adjustment to the machine output. If the machine output and the final human output data structure are completely mismatched, it is directly determined that the human has completely reconstructed the output. If only the local data structure is mismatched, only the heterogeneous part is assigned a value, and the adjustment extent of the remaining structure is calculated normally.

5. The human-computer hierarchical collaborative decision-making and contribution determination method for multidisciplinary educational research according to claim 1, characterized in that: In step S3, the cognitive load index is constructed by comparing the actual human modification time with the historical baseline modification time of similar steps. It is used to characterize the cognitive labor cost borne by humans in correcting machine solutions, thereby amplifying or attenuating the basic contribution value.

6. The human-computer hierarchical collaborative decision-making and contribution determination method for multidisciplinary educational research according to claim 1, characterized in that: In step S3, the time-series constraint index is calculated based on the quality defects of all preceding steps on which the current step depends and the constraint weights of the preceding steps on the current step. It is used to correct the contribution ratio of human deep innovation under poor preceding output.

7. The human-computer hierarchical collaborative decision-making and contribution determination method for multidisciplinary educational research according to claim 1, characterized in that: In stage S4, the chained hash encryption involves concatenating the standardized decision state record of each decision step with the hash digest of the previous step and then encrypting it with SHA-256 to form a serial evidence chain for storage as a traceability credential.

Images