Multi-dimensional credibility passport generation method and system for intelligence decision

Abstract

The invention discloses an intelligence decision-oriented multi-dimensional credibility passport generation method and system, and relates to the related field of artificial intelligence credible computation.The method comprises the steps that target content is preprocessed, metadata is extracted, and the metadata carries a unique content identity label; for the metadata, ternary uncertainty analysis is executed, an analysis result is determined, and analysis dimensions comprise a knowledge traceability dimension, a demonstration structure dimension and an uncertainty dimension; performing credibility comprehensive evaluation on the analysis result, and determining a credibility comprehensive score and a confidence interval; the metadata, the analysis result, the credibility score and the life cycle information are integrated and structuralized into a six-dimensional credibility passport. The technical problem that existing credibility evaluation of the AI generation content lacks a systematic and full-dimensional comprehensive evaluation system adaptive to decision requirements is solved, and the technical effect of constructing a systematic and full-dimensional credibility evaluation system adaptive to different decision scene risk requirements is achieved.

Description

Technical Field

[0001] This application relates to the field of trusted computing in artificial intelligence, and in particular to a method and system for generating multidimensional trustworthy passports for intelligence decision-making. Background Technology

[0002] AI-generated content refers to information products such as text and reports automatically generated by artificial intelligence models. It is widely used in decision support scenarios such as intelligence analysis and market research. The authenticity and reliability of this content directly affect the correctness of decision-making; therefore, its credibility assessment is a crucial step in ensuring decision quality. The reliability of intelligence decisions highly depends on the quality of the credibility assessment of AI-generated content, and its accuracy directly impacts risk control and the final outcome of key decisions. Current mainstream solutions focus on single-dimensional assessment, specifically through methods such as quantifying text fluency using statistical indicators, verifying factual accuracy through knowledge graph comparison, training assessment models with human feedback, and tracing content sources or extracting argumentative structures to analyze logical rationality using blockchain. These methods, because they focus only on a single assessment dimension, lack a unified and integrated assessment framework, and ignore the risk differences in decision-making scenarios and the interpretability and uncertainty of assessment results, result in fragmented assessments, insufficient adaptability, and opaque decision-making basis.

[0003] Currently, the credibility assessment of AI-generated content faces technical challenges due to the lack of a systematic, comprehensive, and decision-making-adaptive evaluation system. Summary of the Invention

[0004] This application provides a multi-dimensional credibility passport generation method and system for intelligence decision-making. It employs preprocessing of target content, extraction of metadata containing unique content identifiers, and a ternary uncertainty analysis of the metadata across three dimensions: knowledge origin, argument structure, and uncertainty. The analysis results are then comprehensively evaluated for credibility, determining a comprehensive score and confidence interval. Finally, metadata, analysis results, credibility scores, and lifecycle information are integrated to structurally generate a six-dimensional credibility passport. This approach addresses the technical problem of existing AI-generated content credibility assessments lacking a systematic, comprehensive, and decision-adaptive comprehensive evaluation system. It achieves the technical effect of constructing a systematic, comprehensive credibility assessment system adapted to the risk requirements of different decision-making scenarios.

[0005] This application provides a method for generating a multi-dimensional credibility passport for intelligence decision-making, comprising: preprocessing target content to extract metadata, wherein the metadata carries a unique content identity identifier; performing a ternary uncertainty analysis on the metadata to determine the analysis result, wherein the analysis dimensions include a knowledge tracing dimension, an argument structure dimension, and an uncertainty dimension; performing a comprehensive credibility assessment on the analysis result to determine a comprehensive credibility score and a confidence interval; and integrating the metadata, analysis result, credibility score, and lifecycle information to structure a six-dimensional credibility passport.

[0006] In a possible implementation, the following processing is performed: a first preprocessing layer is deployed using first-order data cleaning and second-order structuring based on bidirectional semantic disambiguation and dynamic coreference resolution; wherein, the second-order structuring includes disambiguation and alignment of entities, concepts, and reference relationships; a second analysis layer is deployed using a ternary analysis branch based on knowledge tracing dimension, argument structure dimension, and uncertainty dimension; a third evaluation layer is deployed using weighted fusion based on cross-layer attention; and the first preprocessing layer, the second analysis layer, and the third evaluation layer are used as a passport generation module.

[0007] In possible implementations, a second analysis layer is deployed using a ternary analysis branch based on the dimensions of knowledge tracing, argument structure, and uncertainty. The following processing is performed: a tracing analysis branch is built using a breadth-first search guided by a knowledge graph, where the search method is a recursive tracing from surface citations to the tracing scale, performing a three-dimensional quantitative evaluation based on authority, timeliness, and completeness; a structural analysis branch is built using claim-evidence-reasoning triple extraction based on syntactic classification and evaluation using a multimodal logic verification framework, where the multimodal logic verification framework is defined based on logical consistency, sufficiency of evidence, reasoning span, and hypothesis dependence; and an uncertainty analysis branch is built based on the knowledge space mapping mechanism, identifying content coverage blind spots under domain ontology constraints, and using the quantification of the impact strength of key hypotheses on the conclusion as the underlying logic.

[0008] In a possible implementation, the processing steps based on the source tracing analysis branch include the following: identifying explicit and implicit references for the metadata; recursively tracing the explicit and implicit references to determine the source data, wherein the tracing scale is up to the original data source or reaches a preset maximum tracing depth; performing a quantitative scoring based on three dimensions—authority, timeliness, and completeness—on any item in the source data to determine a three-dimensional quantitative score; and adjusting the credibility weight of the corresponding source item by detecting and marking circular reference relationships.

[0009] In a possible implementation, the following processing is performed: in the argument structure dimension analysis based on the structural analysis branch, it is determined whether there are contradictions; if there are contradictions of the same claim, the validity score based on the multimodal logic verification framework is fine-tuned by quantifying the severity of the contradiction, wherein the contradictions include numerical contradictions and semantic contradictions.

[0010] In a possible implementation, the following processing is performed: In the six-dimensional credibility passport, the first dimension includes content identity identifiers and metadata; the second dimension includes knowledge tracing graphs and source quality assessment results; the third dimension includes argument structure graphs and reasoning validity assessment results; the fourth dimension includes uncertainty statements; the fifth dimension includes a comprehensive credibility score and confidence interval; and the sixth dimension includes a full lifecycle record of passport generation, review, use, and verification; wherein, the analysis results include the second, third, and fourth dimensions.

[0011] In a possible implementation, the following processing is performed: In the comprehensive credibility assessment based on the third evaluation layer, cross-branch attention fusion based on the analysis results is performed for the comprehensive indicators to determine the first comprehensive coefficient. The comprehensive indicators include at least evidence quality, reasoning validity, information completeness and internal consistency. The comprehensive coefficient assessment of each comprehensive indicator is completed until the fourth comprehensive coefficient is determined. Weighted calculations are performed on the first comprehensive coefficient to the fourth comprehensive coefficient to determine the comprehensive credibility score.

[0012] In a possible implementation, after determining the overall credibility score and confidence interval, the following processing is performed: a decision risk index is generated based on the risk level of the decision scenario; a credibility threshold and review mechanism are dynamically set based on the decision risk index; and if the overall credibility score of the credibility passport is lower than the threshold according to the credibility threshold and review mechanism, a gap analysis report and improvement suggestions are generated.

[0013] In possible implementations, the following processes are also performed: during the verification stage of the credibility passport, a digital signature is added to the verification record, and a hash chain structure is used to manage the passport versions in a chain; at the generation, verification, and key usage nodes of the credibility passport, the operation sequence is solidified by attaching a trusted timestamp.

[0014] This application also provides a multi-dimensional credibility passport generation system for intelligence decision-making, comprising: a metadata extraction module for extracting metadata by preprocessing target content, wherein the metadata carries a unique content identity identifier; a ternary uncertainty analysis module for performing ternary uncertainty analysis on the metadata to determine the analysis result, wherein the analysis dimensions include a knowledge tracing dimension, an argument structure dimension, and an uncertainty dimension; a credibility comprehensive evaluation module for performing a credibility comprehensive evaluation on the analysis result to determine a credibility comprehensive score and confidence interval; and a credibility passport generation module for integrating the metadata, analysis result, credibility score, and lifecycle information to structure a six-dimensional credibility passport.

[0015] This application proposes a multi-dimensional credibility passport generation method and system for intelligence decision-making. First, the target content is preprocessed to extract metadata, which carries a unique content identity identifier. Next, a ternary uncertainty analysis is performed on the metadata to determine the analysis results. The analysis dimensions include knowledge origination, argumentation structure, and uncertainty. Then, a comprehensive credibility assessment is performed on the analysis results to determine a comprehensive credibility score and confidence interval. Finally, the metadata, analysis results, credibility score, and lifecycle information are integrated to structure a six-dimensional credibility passport. Through this process, the proposed method and system achieve the technical effect of constructing a systematic, multi-dimensional credibility assessment system that adapts to the risk needs of different decision-making scenarios. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0017] Figure 1 This is a flowchart illustrating a multi-dimensional credibility passport generation method for intelligence decision-making, provided as an embodiment of this application.

[0018] Figure 2 This is a schematic diagram of the structure of a multi-dimensional credibility passport generation system for intelligence decision-making, provided in an embodiment of this application.

[0019] Figure labeling: Metadata extraction module 10, ternary uncertainty analysis module 20, credibility comprehensive evaluation module 30, credibility passport generation module 40. Detailed Implementation

[0020] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0021] This application provides a method for generating a multi-dimensional credibility passport for intelligence decision-making, such as... Figure 1 As shown, the method includes: Step S100: By preprocessing the target content, metadata is extracted, wherein the metadata carries a unique content identity identifier.

[0022] Specifically, a format parsing tool is used to standardize the target content, supporting multiple formats such as plain text, Markdown, Word documents, and PDF. Text cleaning is achieved by removing irrelevant information such as headers, footers, and formatting tags. Metadata extraction algorithms are used to extract information such as generation time, author, and version. A content hash fingerprint is calculated using the SHA-256 hash algorithm as the basis for content uniqueness verification. A unique passport identifier is generated using UUID version 4. Simultaneously, a core topic list and keywords are identified through topic extraction, and a summary generation algorithm generates a concise summary, ultimately forming structured metadata containing a unique content identity.

[0023] For example, for a Word document containing a forecast analysis of the global semiconductor market in 2026, the metadata extracted after preprocessing includes a unique identifier with UUID 550e8400-e29b-41d4-a716-446655440000, a SHA-256 hash fingerprint of content a3b2c1d4..., a generation timestamp of 2025-12-02T10:30:00Z, and core information such as the semiconductor market, supply chain, and technology trends.

[0024] Step S200: Perform a ternary uncertainty analysis on the metadata to determine the analysis results. The analysis dimensions include the knowledge tracing dimension, the argument structure dimension, and the uncertainty dimension.

[0025] Specifically, the knowledge tracing dimension identifies explicit and implicit citations through a multi-strategy fusion approach, recursively traces citations using a breadth-first search guided by a knowledge graph, and evaluates source quality using a three-dimensional quantification model. The argument structure dimension classifies sentences using a hybrid rule-based and machine learning approach, extracts claim-evidence-reasoning triples, constructs an argument structure graph, and evaluates the validity of reasoning based on a multimodal logic verification framework. The uncertainty dimension, based on a list of key issues from a domain knowledge base, identifies knowledge gaps through semantic matching, extracts key hypotheses using pattern matching and reasoning chain analysis, and performs sensitivity analysis using a parameterized inference model.

[0026] For example, in the metadata of semiconductor market forecasting analysis, the knowledge tracing dimension identified 15 sources and traced them to 4 levels of depth; the argumentation structure dimension extracted the claim of 35% market growth and the corresponding 12 evidence nodes; and the uncertainty dimension identified the knowledge gap of missing specific data in the Chinese market and the key assumption of 3% global GDP growth.

[0027] Step S300: Perform a comprehensive confidence assessment on the analysis results to determine the comprehensive confidence score and confidence interval.

[0028] Specifically, four comprehensive indicators—evidence quality, reasoning validity, information completeness, and internal consistency—are identified. A weighted adaptive rule system is constructed based on domain knowledge and expert experience, adjusting the weights of each indicator according to content type; for example, reasoning validity has a weight of 0.35 and evidence quality a weight of 0.30 in predictive analysis. A weighted average combined with a non-linear penalty mechanism is used to calculate the comprehensive score, imposing an additional penalty when a certain indicator score falls below a threshold. Confidence interval estimation is used to quantify the uncertainty of the score, including three core sources: data uncertainty is measured by the variance of the evidence source quality score and the variance of the reasoning step validity score; sample size uncertainty is penalized by the number of pieces of evidence—the smaller the sample size, the larger the penalty coefficient (e.g., 0.2 for ≤3 pieces of evidence, 0.1 for 3-5 pieces, and 0 for ≥5 pieces); and model uncertainty is estimated by evaluating the variance of the model's prediction results through 5-fold cross-validation or integrating three different architectures. The comprehensive uncertainty is synthesized using an error propagation formula, first calculating the variances of the three types of uncertainty, then summing the variances and taking the square root to obtain the comprehensive standard error. The confidence interval is constructed based on the assumption of normal distribution. The formula is the mean of the comprehensive score ± the critical value of the corresponding confidence level × the comprehensive standard error. The critical values ​​are 1.645 for 90% confidence level, 1.96 for 95% confidence level, and 2.576 for 99% confidence level. The confidence level can be flexibly adjusted according to the decision risk preference.

[0029] For example, the overall score for semiconductor market forecast analysis is 0.81, with data uncertainty variance of 0.0004, sample size uncertainty variance of 0.0002, and model uncertainty variance of 0.0003. The overall standard error is √(0.0004+0.0002+0.0003)=0.03, and the 95% confidence interval is 0.81±1.96×0.03, i.e. [0.75,0.87]. If the decision is a high-risk investment decision, it can be adjusted to a 99% confidence level, corresponding to an interval of [0.73,0.89].

[0030] Step S400: Integrate the metadata, analysis results, credibility score, and lifecycle information to structure a six-dimensional credibility passport. The six dimensions of the credibility passport include: the first dimension containing content identity and metadata; the second dimension containing a knowledge tracing graph and source quality assessment results; the third dimension containing an argument structure graph and reasoning validity assessment results; the fourth dimension containing an uncertainty statement; the fifth dimension containing a comprehensive credibility score and confidence interval; and the sixth dimension containing a complete lifecycle record of passport generation, review, use, and verification. The analysis results include the second, third, and fourth dimensions.

[0031] Specifically, following a pre-defined six-dimensional structured data model, data modules for each dimension are linked through standard interfaces. The first dimension integrates content identity identifiers and metadata; the second encapsulates knowledge tracing graphs and source quality assessment results; the third inputs argument structure graphs and reasoning validity assessment results; the fourth organizes uncertainty statement information; the fifth records comprehensive credibility scores and confidence intervals; and the sixth collects data throughout the entire lifecycle of passport generation, review, use, and verification. Structured storage in JSON format ensures the relevance and scalability of data across modules.

[0032] In one possible implementation, the method further includes step S500, deploying a first preprocessing layer with first-order data cleaning and second-order structured processing based on bidirectional semantic disambiguation and dynamic coreference resolution. The second-order structured processing includes disambiguation and alignment of entities, concepts, and reference relationships. Specifically, first-order data cleaning uses regular expressions to remove redundant formatting tags and special characters, and divides the content into logically coherent paragraphs using a text segmentation algorithm. In the second-order structured processing, entity disambiguation uses a BERT fine-tuning model based on contextual semantic similarity to distinguish entities with the same name; concept disambiguation achieves standardized alignment by comparing with a domain ontology; and reference relationship disambiguation combines contextual information such as publication time and organization name to distinguish sources with the same name. Dynamic coreference resolution uses an attention-based referencing resolution model to identify the correspondence between pronouns and antecedents in the text.

[0033] For example, when dealing with texts about the competition between Huawei and Apple in the 5G field, entity disambiguation clarifies that Apple is a technology company, reference disambiguation distinguishes between the Apple report released in 2024 and the Apple report released in 2023, and common reference analysis determines that "its" in "its technological advantages" refers to Huawei.

[0034] Step S600 involves deploying the second analysis layer using a ternary analysis branch based on the dimensions of knowledge tracing, argument structure, and uncertainty. Specifically, a parallel architecture is used to deploy three independent analysis branches. Each branch receives structured data output from the preprocessing layer through a standardized data interface, independently completes its analysis, and outputs the results to the evaluation layer. The tracing analysis branch uses a knowledge graph query engine for recursive tracing, the structure analysis branch performs structure deconstruction and evaluation through argument mining, and the uncertainty analysis branch performs relevant analysis using a domain knowledge base and sensitivity analysis algorithms. A data synchronization mechanism is established between the branches to ensure data consistency during cross-branch contradiction detection.

[0035] For example, when three branches process a new energy vehicle market analysis report simultaneously, the tracing branch traces the source while the structural branch extracts the argumentation relationship, and the uncertainty branch identifies knowledge gaps. The parallel processing of these three branches improves the overall analysis efficiency by 40%.

[0036] Step S700 involves deploying the third evaluation layer using weighted fusion based on cross-layer attention. Specifically, a cross-layer attention mechanism model is constructed, taking the outputs of the three branches of the second analysis layer as input. By calculating the attention weights of the output features of each branch with the comprehensive evaluation index, the contribution of highly correlated features is highlighted. A multilayer perceptron is used to perform a nonlinear transformation on the fused features, and coefficients for each comprehensive index are generated by combining them with preset scoring rules. A feedback adjustment mechanism is introduced to dynamically optimize the attention weight calculation parameters based on the correlation between historical evaluation data and actual decision results.

[0037] Step S800: The first preprocessing layer, the second analysis layer, and the third evaluation layer are used as a passport generation module. Specifically, a modular architecture design is adopted, and communication and data transmission between the three layers are realized through API interfaces. The output of the preprocessing layer serves as the input of the analysis layer, the results of the analysis layer are passed to the evaluation layer, and the output of the evaluation layer is integrated with lifecycle information to generate a passport. The module has a built-in load balancing mechanism to support multi-task parallel processing, and adjusts the computing resources of each layer according to the task complexity through a dynamic resource allocation algorithm. A configurable interface is provided, allowing users to adjust parameters such as preprocessing rules, analysis depth, and evaluation weights according to actual needs.

[0038] In one possible implementation, a second analysis layer is deployed using a ternary analysis branch based on the dimensions of knowledge tracing, argumentation structure, and uncertainty. Step S600 further includes step S610, which establishes a tracing analysis branch using a breadth-first search guided by a knowledge graph. The search method is a recursive tracing from surface citations to the tracing scale, performing a three-dimensional quantitative evaluation based on authority, timeliness, and completeness. Specifically, a knowledge graph is constructed containing academic literature, industry reports, official data, etc., storing attributes such as source identifiers, types, publication times, and citation relationships. A breadth-first search algorithm is used, starting from surface citations in the content, prioritizing the tracing of upstream sources with high citation frequency, recursively tracing to the original data source or reaching a preset maximum tracing depth. The original data source includes official data from the National Bureau of Statistics, original experimental records, etc. The maximum tracing depth is set according to the content type; for example, academic content defaults to 4 layers, and news content defaults to 2 layers. The authority assessment is based on a predefined database of institutional authority, including journal impact factors, research institution rankings, and government department levels. The timeliness assessment calculates the time elapsed since publication and considers the field's update cycle. The completeness assessment checks whether the source includes key information such as research methods, raw data, and peer review records. A weighted average is used to calculate the overall quality score across the three dimensions, with the weights adaptively adjusted according to the characteristics of the field.

[0039] Step S620 involves extracting claim-evidence-reasoning triplets based on syntactic classification and evaluating them using a multimodal logic verification framework to build a structural analysis branch. The multimodal logic verification framework is defined based on logical consistency, sufficiency of evidence, reasoning span, and hypothesis dependency. Specifically, syntactic classification employs a fusion of rule-based classifiers and machine learning classifiers. The rule-based classifier identifies sentence types based on linguistic features. For example, claim sentences typically contain assertive words such as "will," "expect," "indicate," and "prove," and are often declarative sentences without further support from subsequent sentences. Evidence sentences typically contain data, citation markers, expert opinion introductory words, and case descriptions. Reasoning sentences contain logical connectors such as "because," "therefore," "thus," "based on," "can be inferred," and "given," connecting the claim and evidence. Sentences can be further subdivided, such as claims being categorized as factual claims (stating objective facts or data), predictive claims (predictions about the future), and value judgments (evaluations of good or bad). Evidence is categorized as quantitative data (statistics), qualitative opinions (expert opinions), and case evidence (specific examples). Reasoning is categorized into inductive reasoning (from specific to general), deductive reasoning (from general to specific), causal reasoning (causal relationship arguments), and analogical reasoning (analogical arguments). A machine learning classifier, trained on a fine-tuned BERT model on labeled corpora, captures complex patterns that rules struggle to cover. The results of the two classifiers are fused; the rule-based classifier offers high accuracy, while the machine learning classifier provides high recall, and their combination improves overall performance. Logical relationships between sentences are determined through argument relation extraction. Dependency analysis and discourse relation identification are used to extract triples, identifying which evidence supports which claims and which inferences connect evidence and claims, thus constructing a directed acyclic graph (DAG) of arguments, where nodes represent sentences and edges represent argument relations. The DAG undergoes legality checks, including acyclicity verification, connectivity checks, and isolated node detection. Topological sorting is used for acyclicity verification, connectivity checks ensure that main claims are supported by chains of evidence, and isolated node detection identifies uncited evidence or unsupported claims.

[0040] A multimodal logic verification framework is used to evaluate the validity of reasoning, comprising four dimensions. Logical consistency checks whether the reasoning conforms to basic logical rules, using a formal logic verifier to check the validity of structural deductive reasoning such as syllogisms, and checking for obvious causal inversions or confusion between relevance and causality in causal reasoning. Evidence sufficiency assesses the adequacy of evidence supporting the claim, including the quantity, diversity, and relevance of evidence to the claim. Single pieces of evidence are insufficient; multiple independent pieces of evidence enhance credibility; evidence from different types and sources is more reliable. Reasoning span assesses the logical distance from evidence to conclusion, using semantic vectors to calculate the semantic similarity between evidence and conclusion. Excessive semantic distance indicates logical jumps, potentially missing intermediate reasoning steps. Hypothesis dependence identifies implicit assumptions in the reasoning, assessing the reasonableness of these assumptions and their impact on the conclusion. Reasoning heavily reliant on unreasonable assumptions has low validity. The four dimensions are combined to score the validity of the reasoning using the weakest link effect; a low score in one dimension significantly lowers the overall score, even if other dimensions score highly, reflecting the characteristic that a single error in logical reasoning can lead to overall failure. For example, a certain reasoning might have a logical consistency score of 0.90, an evidence sufficiency score of 0.85, and a reasoning span score of 0.80, but a hypothesis dependence score of only 0.30. As a result, the final reasoning validity score is significantly lowered to 0.55, rather than the simple average of the four dimensions of 0.71.

[0041] Step S630: Based on the knowledge space mapping mechanism, content coverage blind spots are identified under the constraints of domain ontology. An uncertainty analysis branch is built using the quantification of the impact of key assumptions on the conclusions as the underlying logic. Specifically, an ontology library for each domain and a list of key questions corresponding to each topic are constructed. The list includes the core dimensions that the topic should cover, such as market size and growth trends in market analysis. The topic domain of the input content is identified through a topic classification model, and the corresponding list of key questions is retrieved. Semantic matching technology is used to determine the coverage of each key question. Coverage is calculated based on the length, depth, and clarity of relevant paragraphs. Questions with low coverage but high importance are marked as knowledge gaps and categorized into three severity levels—high, medium, and low—based on their importance and degree of coverage deficiency. Supplementary suggestions are generated for each knowledge gap, specifying the type of missing information and the source of recommended information. Simultaneous temporal and geographic gap detection is performed. Temporal gap detection extracts the publication time or coverage period of cited data within the content, calculates the elapsed time, and compares it with typical update cycles in the domain. Data exceeding a reasonable timeframe is marked as temporal gaps, and the impact of outdated data on conclusions is assessed. Geographical gap detection extracts mentioned countries and regions through named entity recognition and compares them with a list of key geographic regions for the topic; data lacking important regions is marked as geographic gaps. Key hypothesis extraction is divided into explicit and implicit categories. Explicit hypotheses are captured through pattern matching to identify expressions such as "hypothesis…", "assume…", "premise…", and "under…conditions." Implicit hypotheses are extracted through reasoning chain analysis combined with causal reasoning models and common sense knowledge bases, identifying unstated but necessary preconditions in the reasoning process. A criticality rating is applied to the extracted hypotheses, defining reasonable value ranges or variation scenarios for each hypothesis. Reasoning is then repeated under different hypothesis values, and the magnitude of change in conclusions is observed, resulting in high, medium, and low criticality levels. Sensitivity analysis is performed on highly critical hypotheses. A parametric inference model is constructed, with the hypotheses as input parameters and the conclusions as outputs. The range of change in the conclusions is calculated within a reasonable range of change in the hypotheses. The sensitivity index is obtained by dividing the magnitude of change in the conclusions by the magnitude of change in the hypotheses. The higher the value, the more significant the impact.

[0042] In one possible implementation, in the source tracing analysis branch processing steps, step S610 further includes step S611, which identifies explicit and implicit citations for the metadata. Specifically, explicit citation identification uses regular expressions to match academic citation formats, report citation formats, and web citation formats. Academic citation formats include author-year, numbered citations, etc.; report citation formats include report name-publishing institution, etc.; web citation formats include URL, DOI, etc. Implicit citation identification identifies lead words such as "according to...", "research shows...", "data shows...", and extracts source institution and time information by combining named entity recognition technology. A rule-based filtering algorithm is used to remove false citations, such as statements without actual source references. Threshold filtering is used to ensure the accuracy of citation identification, such as citation-related words appearing at a frequency ≥2.

[0043] Step S612 involves recursively tracing the explicit and implicit references to determine the source data, with the tracing scale being the original data source or reaching a preset maximum tracing depth. Specifically, based on the knowledge graph query interface, a query is initiated for each identified reference to obtain its upstream source. A breadth-first strategy is used to trace the reference relationships of the upstream sources sequentially, recording the depth of each tracing path. A tracing termination condition is set, stopping when the original data source is reached or the preset maximum tracing depth is reached. The source information obtained during the tracing process is structured and stored to form source data containing source identifier, type, publication time, tracing path, etc.

[0044] Step S613: Perform a quantitative scoring based on three dimensions—authority, timeliness, and completeness—on any item in the traceability data to determine the three-dimensional quantitative score. Specifically, the credibility weight of the corresponding traceability item is adjusted by detecting and marking circular citation relationships. The authority score uses a 0-1 point scale, calculated by weighting the inherent authority of the source type and the institution's reputation rating. For example, academic journals score 0.8-1.0, online articles 0.2-0.4, and Top 10 research institutions 0.9-1.0. The timeliness score is calculated inversely based on the ratio of the time since publication to the field's update cycle. For example, score = 1 - (time since publication ÷ field update cycle). If the time since publication exceeds the update cycle, the score is 0. For example, in the science and technology field, the update cycle is 6 months; a source published 3 months ago receives 0.5 points, and a source published 12 months ago receives 0 points. The completeness score checks whether the source contains key information such as research methods, original data, and peer review records; the percentage of items meeting these criteria is the score. The Tarjan algorithm is used to construct a directed graph of references, detect strongly connected components to identify circular references, and mark them as information echo chamber risks, reducing the credibility weight of the corresponding traceability item by 30%-50%. The three-dimensional quantitative scoring calculates the comprehensive score through weighted average, with each weight accounting for 1 / 3 by default, but this can be adaptively configured.

[0045] In one possible implementation, in the structural analysis of the argument structure dimension based on the structural analysis branch, step S620 further includes step S621, determining whether there are contradictions. If contradictions exist for the same claim, the validity score based on the multimodal logic verification framework is fine-tuned by quantifying the severity of the contradiction. The contradictions include numerical contradictions and semantic contradictions. Specifically, a pairwise comparison method is used to examine all evidence supporting the same claim. Numerical contradiction detection extracts numerical statements from the evidence, such as growth rates, calculates the numerical difference rate, and compares it with a preset reasonable difference threshold. The difference threshold is set according to the domain; for example, the default is 5% for economic data. Exceeding the threshold indicates a numerical contradiction. Semantic contradiction detection uses a natural language reasoning model, such as a fine-tuned RoBERTa model, to determine whether there is a semantic opposition between two pieces of evidence, such as market prosperity versus market contraction. A confidence level ≥ 0.8 indicates a semantic contradiction. The severity of the contradiction is quantified into different levels, such as slight, moderate, and severe, each corresponding to a deduction of different validity scores.

[0046] In one possible implementation, in the credibility comprehensive assessment based on the third evaluation layer, step S700 further includes step S710, which, for the comprehensive indicators, performs cross-branch attention fusion based on the analysis results to determine the first comprehensive coefficient. The comprehensive indicators at least include evidence quality, reasoning validity, information completeness, and internal consistency. Specifically, the core dimensions of the comprehensive indicators strictly correspond to intelligence decision-making needs. The evidence quality dimension integrates the results of knowledge tracing analysis, with core indicators including statistical characteristics of source authority (mean, median, minimum), source timeliness, source completeness, source diversity (proportion of different types of sources), and evidence independence (proportion of independent sources). The reasoning validity dimension integrates the results of argumentation structure analysis, with core indicators including statistical characteristics of the four-dimensional scores of reasoning steps (logical consistency, evidence sufficiency, reasoning span, hypothesis dependence). The information completeness dimension assesses actual coverage, the proportion of different viewpoints, data time span, and geographical coverage based on a list of key issues defined in the domain knowledge base. The internal consistency dimension focuses on checking whether the claims in different parts of the content are contradictory, whether the same data is consistently expressed, and whether the conclusion corresponds to the preceding analysis, directly linking to the contradiction detection results. The cross-branch attention fusion model takes the outputs of three analysis branches as input features. It calculates the correlation score between each feature and its corresponding comprehensive index, normalizes the scores using the Softmax function to obtain attention weights, and then multiplies the features by the weights and sums the results to obtain the first comprehensive coefficient. Model training uses the mean squared error between the comprehensive coefficient and the manually labeled results as the loss function, iteratively optimizing parameters to improve matching accuracy.

[0047] Step S720 involves evaluating the comprehensive coefficients of each comprehensive indicator until the fourth comprehensive coefficient is determined. Weighting calculations are then performed on the first through fourth comprehensive coefficients to determine the overall credibility score. Specifically, following the cross-branch attention fusion logic of step S710, the inference validity (second comprehensive coefficient), information completeness (third comprehensive coefficient), and internal consistency (fourth comprehensive coefficient) are calculated sequentially, ensuring that each coefficient is generated based on the core indicators of its corresponding dimension. The weighting calculation employs an adaptive weighting mechanism. Weights are set based on a rule system encoded with domain knowledge and expert experience, dynamically adjusted according to content type. Predictive analysis emphasizes inference validity and hypothesis reasonableness, factual reporting emphasizes evidence quality and information completeness, and comparative analysis increases the weight of information completeness and internal consistency. The overall score calculation uses a weighted average combined with a non-linear penalty mechanism. If any key dimension coefficient falls below a preset threshold, such as 0.6, an additional penalty is applied to the overall score. The penalty intensity increases non-linearly as the coefficient decreases, reflecting the weakest link effect.

[0048] For example, in a predictive analysis, the first comprehensive coefficient (quality of evidence) is 0.80, the second comprehensive coefficient (validity of reasoning) is 0.85, the third comprehensive coefficient (completeness of information) is 0.75, and the fourth comprehensive coefficient (internal consistency) is 0.82, with corresponding weights of 0.25, 0.35, 0.2, and 0.2, respectively. The weighted sum yields an initial comprehensive score of 0.81. Since no coefficient falls below the threshold, the final credibility comprehensive score is 0.81. Conversely, if a factual report has an evidence quality coefficient of 0.55, which is below the threshold, the initial weighted sum is 0.78. After a penalty of 0.8, the final comprehensive score is 0.62.

[0049] In one possible implementation, after determining the comprehensive credibility score and confidence interval, the method further includes step S900: generating a decision risk index based on the risk level of the decision-making scenario. Specifically, a five-dimensional quantitative model of the decision risk index is constructed. The economic impact dimension considers the direct and indirect economic value involved in the decision, normalizing investment amount, potential returns, and possible losses to the 0-1 range, and using a logarithmic scale to process data across orders of magnitude. The time urgency dimension assesses the length of the decision window, assigning values ​​based on the window length; the more urgent the time, the higher the risk, for example, ≤1 week = 1.0, 1-4 weeks = 0.7, and more than 4 weeks = 0.3. The irreversibility dimension assesses the degree of decision revocability; completely irreversible decisions have higher risk than reversible decisions. The decision risk index is assigned based on the degree of revocability, such as 1.0 for completely irreversible, 0.5 for partially reversible, and 0.1 for completely reversible. The scope of impact considers the number of people or organizations affected by the decision; the wider the scope, the greater the responsibility. It is assigned a value based on the number of affected people / organizations, such as 1.0 for ≥1000 people, 0.7 for 100-1000 people, and 0.3 for <100 people. The sensitivity dimension assesses whether the decision involves socially sensitive issues; highly sensitive decisions require greater caution. It is assigned a value based on the degree of sensitivity, such as 1.0 for high sensitivity, 0.5 for medium sensitivity, and 0.1 for low sensitivity. A weighted average decision risk index is calculated, ranging from 0 to 1, with higher values ​​indicating higher risk. The default weight for each value is 0.2, which can be adjusted according to the organization's risk appetite.

[0050] Step S1000: Based on the decision risk index, dynamically set the credibility threshold and review mechanism. Specifically, construct a credibility threshold adjustment function. High-risk decisions / decision risk indices require higher credibility thresholds, narrower confidence intervals, and more manual review; low-risk decisions / decision risk indices can accept lower credibility thresholds to reduce unnecessary verification costs. The credibility threshold adjustment function is set as a monotonically increasing function of the decision risk index, such as threshold = 0.5 + 0.5 × decision risk index. The review mechanism is set as follows: low-risk decisions use fully automatic review; medium-risk decisions use automatic review and manual spot checks (30% spot check ratio); high-risk decisions use automatic review, full manual review, and expert review. Simultaneously, a confidence interval width requirement is set. High-risk decisions require not only high credibility but also low uncertainty, i.e., a narrow confidence interval, such as low-risk decisions allowing a width ≤ 0.08, medium-risk ≤ 0.05, and high-risk ≤ 0.03.

[0051] Step S1100: Based on the credibility threshold and review mechanism, if the overall credibility score of the credibility passport is lower than the threshold, a gap analysis report and improvement suggestions are generated. Specifically, when the overall credibility score of the credibility passport is lower than the threshold, the gap between each comprehensive indicator and its corresponding score threshold is analyzed. For example, the evidence quality score threshold is 0.8, the actual score is 0.72, and the gap is 0.08. The gap analysis report includes the name of the indicator that does not meet the requirements, its current value, the required value, the size of the gap, and the reason for the gap. Improvement suggestions are generated based on the reasons for the gap, specifying specific action plans, responsible parties, and completion deadlines, such as supplementing specific types of evidence sources, conducting manual review to correct reasoning logic, and updating outdated data. The report is presented in a structured report format and supports exporting to Word or PDF files.

[0052] In one possible implementation, the method further includes step S1200: during the verification stage of the credible passport, a digital signature is added to the verification record, and a hash chain structure is used for chained management of passport versions. Specifically, when the verifier logs into the system, they use identity authentication, such as an account password and dynamic password. After the verification is completed, the system automatically calculates the hash value of the verification record, and the verifier uses their private key to sign the hash value, generating an unrepudiable verification certificate. Passport version management uses hash chain technology. Each version calculates a content hash value as a version fingerprint, and the new version records the hash value of the previous version, forming a chain structure. The system stores the modified content, modification time, modifier, and reason for modification for each version, supports version backtracking query, and any modification to a historical version will cause the hash value of subsequent versions to change, thus being detected.

[0053] Step S1300: At key nodes in the generation, review, and use of the trusted passport, the operation sequence is solidified by attaching a trusted timestamp. Specifically, a third-party timestamp authentication service is integrated. At key nodes such as passport generation completion, approval, first use, and major version updates, requests are sent to the third-party timestamp authentication service to obtain a trusted timestamp containing the precise time and the digital signature of the third-party timestamp authentication service. The timestamp information is bound and stored with the corresponding operation record, including operation type, operator, operation time, and timestamp data. The system supports timestamp verification. Users can verify the authenticity and integrity of the timestamp through the verification interface provided by the third-party timestamp authentication service, ensuring that the operation sequence cannot be tampered with.

[0054] This application employs preprocessing of the target content, extracting metadata containing unique content identifiers, and conducting ternary uncertainty analysis on the metadata from three dimensions: knowledge tracing, argumentation structure, and uncertainty. The analysis results are then used to conduct a comprehensive credibility assessment, determining the comprehensive score and confidence interval. Furthermore, by integrating metadata, analysis results, credibility scores, and lifecycle information, a six-dimensional credibility passport is generated in a structured manner. These technical means address the technical problem of existing AI-generated content credibility assessments lacking a systematic, comprehensive, and decision-adaptive comprehensive assessment system. This achieves the technical effect of constructing a systematic, comprehensive credibility assessment system that adapts to the risk requirements of different decision-making scenarios.

[0055] In the above text, refer to Figure 1 A multi-dimensional credibility passport generation method for intelligence decision-making according to an embodiment of the present invention is described in detail. Next, reference will be made to... Figure 2 This invention describes a multi-dimensional credibility passport generation system for intelligence decision-making according to an embodiment of the present invention.

[0056] A multi-dimensional credibility passport generation system for intelligence decision-making, according to an embodiment of the present invention, addresses the technical problem of the lack of a systematic, comprehensive, and decision-adaptive evaluation system for the credibility assessment of existing AI-generated content. It achieves the technical effect of constructing a systematic, comprehensive credibility assessment system adapted to the risk requirements of different decision-making scenarios. The multi-dimensional credibility passport generation system for intelligence decision-making includes: a metadata extraction module 10, a ternary uncertainty analysis module 20, a credibility comprehensive assessment module 30, and a credibility passport generation module 40.

[0057] Metadata extraction module 10 is used to extract metadata by preprocessing the target content, wherein the metadata carries a unique content identity identifier; ternary uncertainty analysis module 20 is used to perform ternary uncertainty analysis on the metadata to determine the analysis result, wherein the analysis dimensions include knowledge tracing dimension, argument structure dimension, and uncertainty dimension; credibility comprehensive evaluation module 30 is used to perform credibility comprehensive evaluation on the analysis result to determine the credibility comprehensive score and confidence interval; credibility passport generation module 40 is used to integrate the metadata, analysis result, credibility score, and lifecycle information to structure a six-dimensional credibility passport.

[0058] The system may further include: a first preprocessing layer deployment module for deploying a first preprocessing layer using first-order data cleaning and second-order structured processing based on bidirectional semantic disambiguation and dynamic coreference resolution, wherein the second-order structured processing includes disambiguation and alignment of entities, concepts, and reference relationships; a second analysis layer deployment module for deploying a second analysis layer using a ternary analysis branch based on knowledge tracing dimension, argument structure dimension, and uncertainty dimension; a third evaluation layer deployment module for deploying a third evaluation layer using weighted fusion based on cross-layer attention; and a passport generation module for using the first preprocessing layer, the second analysis layer, and the third evaluation layer as a passport generation module.

[0059] The second analysis layer is deployed using a ternary analysis branch based on the dimensions of knowledge tracing, argument structure, and uncertainty. The second analysis layer deployment module can further include: a tracing analysis branch building unit for building a tracing analysis branch using a breadth-first search guided by a knowledge graph, wherein the search method is a recursive tracing from surface citations to the tracing scale, performing a three-dimensional quantitative evaluation based on authority, timeliness, and completeness; a structure analysis branch building unit for building a structure analysis branch using claim-evidence-reasoning triple extraction based on syntactic classification and evaluation using a multimodal logic verification framework, wherein the multimodal logic verification framework is defined based on logical consistency, sufficiency of evidence, reasoning span, and hypothesis dependence; and an uncertainty analysis branch building unit for identifying content coverage blind spots under domain ontology constraints based on a knowledge space mapping mechanism, using the quantification of the influence of key hypotheses on the conclusion as the underlying logic to build an uncertainty analysis branch.

[0060] The source tracing analysis branch construction unit can further include: a reference identification subunit for identifying explicit and implicit references to the metadata; a recursive tracing subunit for recursively tracing the explicit and implicit references to determine the source data, wherein the tracing scale is up to the original data source or reaches a preset maximum tracing depth; and a three-dimensional quantitative scoring subunit for performing a quantitative scoring based on three dimensions—authority, timeliness, and completeness—on any item in the source data to determine the three-dimensional quantitative score, wherein the credibility weight of the corresponding source item is adjusted by detecting and marking circular reference relationships.

[0061] In the structural analysis of the argumentation structure dimension based on the structural analysis branch, the structural analysis branch construction unit may further include: a contradiction point determination subunit for determining whether there are contradiction points. If there are contradiction points of the same claim, the validity score based on the multimodal logic verification framework is fine-tuned by quantifying the severity of the contradiction. The contradiction points include numerical contradictions and semantic contradictions.

[0062] The detailed description of the specific configuration of the credibility passport generation module 40 is explained as follows: As mentioned above, the credibility passport generation module 40 may further include: in the six-dimensional credibility passport, the first dimension includes content identity identifier and metadata, the second dimension includes knowledge tracing graph and source quality assessment results, the third dimension includes argument structure graph and reasoning validity assessment results, the fourth dimension includes uncertainty statement, the fifth dimension includes credibility comprehensive score and confidence interval, and the sixth dimension includes the full life cycle record of passport generation, review, use and verification, wherein the analysis results include the second, third and fourth dimensions.

[0063] In the credibility comprehensive evaluation based on the third evaluation layer, the third evaluation layer deployment module may further include: a cross-branch attention fusion unit for performing cross-branch attention fusion based on the analysis results for comprehensive indicators to determine the first comprehensive coefficient, wherein the comprehensive indicators include at least evidence quality, reasoning validity, information completeness and internal consistency; and a weighting calculation unit for completing the comprehensive coefficient evaluation of each comprehensive indicator until the fourth comprehensive coefficient is determined, performing weighting calculation on the first comprehensive coefficient up to the fourth comprehensive coefficient to determine the credibility comprehensive score.

[0064] After determining the overall credibility score and confidence interval, the system may further include: a decision risk index generation module for generating a decision risk index based on the risk level of the decision scenario; a credibility threshold dynamic setting module for dynamically setting a credibility threshold and review mechanism based on the decision risk index; and a gap analysis report generation module for generating a gap analysis report and improvement suggestions if the overall credibility score of the credibility passport is lower than the threshold, based on the credibility threshold and review mechanism.

[0065] The system may further include: a chain management module for adding digital signatures to the review records during the review stage of the credible passport, and using a hash chain structure to manage the passport versions in a chain; and an operation sequence solidification module for solidifying the operation sequence by attaching a trusted timestamp at the generation, review and key usage nodes of the credible passport.

[0066] The multi-dimensional credibility passport generation system for intelligence decision-making provided in this embodiment of the invention can execute the multi-dimensional credibility passport generation method for intelligence decision-making provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0067] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0068] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for generating a multi-dimensional credibility passport for intelligence decision-making, characterized in that, The method includes: Metadata is extracted by preprocessing the target content, wherein the metadata carries a unique content identity identifier; For the aforementioned metadata, a ternary uncertainty analysis is performed to determine the analysis results, wherein the analysis dimensions include the knowledge origination dimension, the argumentation structure dimension, and the uncertainty dimension; A comprehensive confidence assessment is performed on the analysis results to determine the comprehensive confidence score and confidence interval; The metadata, analysis results, credibility scores, and lifecycle information are integrated and structured into a six-dimensional credibility passport.

2. The multi-dimensional credibility passport generation method for intelligence decision-making as described in claim 1, characterized in that, The method includes: The first preprocessing layer is deployed by first-order data cleaning and second-order structured processing based on bidirectional semantic disambiguation and dynamic coreference resolution. The second-order structured processing includes disambiguation and alignment of entities, concepts, and reference relationships; The second analysis layer is deployed using a ternary analysis approach based on the dimensions of knowledge tracing, argumentation structure, and uncertainty. A third evaluation layer is deployed using weighted fusion based on cross-layer attention; The first preprocessing layer, the second analysis layer, and the third evaluation layer are used as the passport generation module.

3. The multi-dimensional credibility passport generation method for intelligence decision-making as described in claim 2, characterized in that, The second analytical layer is deployed using a ternary analytical approach based on the dimensions of knowledge origination, argumentation structure, and uncertainty, including: A knowledge graph-guided breadth-first search is used to build a source tracing analysis branch. The search method is a recursive tracing from surface citations to the traceability scale, and a three-dimensional quantitative evaluation based on authority, timeliness and completeness is performed. We construct a structural analysis branch by extracting claim-evidence-reasoning triplets based on syntactic classification and evaluating them with a multimodal logic verification framework. The multimodal logic verification framework is defined based on logical consistency, sufficiency of evidence, reasoning span, and hypothesis dependence. Based on the knowledge space mapping mechanism, content coverage blind spots are identified under the constraints of domain ontology. An uncertainty analysis branch is built with the underlying logic of quantifying the influence of key assumptions on the conclusions.

4. The multi-dimensional credibility passport generation method for intelligence decision-making as described in claim 3, characterized in that, The processing steps based on source tracing analysis branches include: For the aforementioned metadata, identify explicit and implicit references; The explicit and implicit references are recursively traced to determine the source data, wherein the tracing scale is up to the original data source or reaches the preset maximum tracing depth. For any item in the traceability data, a quantitative score based on three dimensions—authority, timeliness, and completeness—is performed to determine the three-dimensional quantitative score; Specifically, by detecting and marking circular references, the credibility weight of the corresponding tracing item is adjusted.

5. The multi-dimensional credibility passport generation method for intelligence decision-making as described in claim 3, characterized in that, In the structural dimension analysis of the argument based on the structural analysis branch, it is determined whether there are contradictions. If there are contradictory points in the same claim, the validity score based on the multimodal logic verification framework is fine-tuned by quantifying the severity of the contradiction. The contradictory points include numerical contradictions and semantic contradictions.

6. The multi-dimensional credibility passport generation method for intelligence decision-making as described in claim 1, characterized in that, The six-dimensional credibility passport includes the following dimensions: the first dimension contains content identity identifiers and metadata; the second dimension contains knowledge tracing graphs and source quality assessment results; the third dimension contains argumentation structure graphs and reasoning validity assessment results; the fourth dimension contains uncertainty statements; the fifth dimension contains a comprehensive credibility score and confidence interval; and the sixth dimension contains a full lifecycle record of passport generation, review, use, and verification. The analysis results include a second dimension, a third dimension, and a fourth dimension.

7. The multi-dimensional credibility passport generation method for intelligence decision-making as described in claim 2, characterized in that, In the credibility comprehensive evaluation based on the third evaluation layer, cross-branch attention fusion based on the analysis results is performed on the comprehensive indicators to determine the first comprehensive coefficient. The comprehensive indicators include at least the quality of evidence, the validity of reasoning, the completeness of information, and internal consistency. Complete the comprehensive coefficient evaluation of each comprehensive indicator until the fourth comprehensive coefficient is determined. Perform weighted calculation on the first comprehensive coefficient up to the fourth comprehensive coefficient to determine the comprehensive credibility score.

8. The multi-dimensional credibility passport generation method for intelligence decision-making as described in claim 1, characterized in that, After determining the overall credibility score and confidence interval, the following is included: Generate a decision risk index based on the risk level of the decision-making scenario; Based on the aforementioned decision risk index, a credibility threshold and review mechanism are dynamically set. Based on the aforementioned credibility threshold and review mechanism, if the overall credibility score of the credibility passport is lower than the threshold, a gap analysis report and improvement suggestions will be generated.

9. The method for generating a multi-dimensional credibility passport for intelligence decision-making as described in claim 1, characterized in that, The method further includes: During the verification stage of the credible passport, digital signatures are added to the verification records, and a hash chain structure is used to manage the passport versions in a chain. At key points in the generation, review, and use of the credibility passport, the operation sequence is solidified by attaching a trusted timestamp.

10. A multi-dimensional credibility passport generation system for intelligence decision-making, characterized in that, The system is used to implement the multi-dimensional credibility passport generation method for intelligence decision-making as described in any one of claims 1-9, the system comprising: The metadata extraction module is used to extract metadata by preprocessing the target content, wherein the metadata carries a unique content identity identifier; The ternary uncertainty analysis module is used to perform ternary uncertainty analysis on the metadata and determine the analysis results. The analysis dimensions include the knowledge tracing dimension, the argument structure dimension, and the uncertainty dimension. The credibility comprehensive evaluation module is used to perform a credibility comprehensive evaluation on the analysis results and determine the credibility comprehensive score and confidence interval; The credibility passport generation module is used to integrate the metadata, analysis results, credibility scores and lifecycle information, and structure them into a six-dimensional credibility passport.

Images