Method for constructing consensus of group risk cognition based on risk metaphor

Abstract

The application relates to the technical field of natural language processing, and particularly discloses a group risk cognitive consensus construction method based on risk metaphor, multi-source interactive texts are collected, and after extraction and filtration, an initial metaphor cognitive set is formed; a dynamic relationship network is constructed based on the set, responsibility is quantified as an attribute quantity transmitted between nodes, and a responsibility distribution graph is generated through iterative calculation; the transmission path is calculated, the dispersion and backtracking features thereof are quantified, and a quantitative analysis report for identifying a high-risk responsibility dilution path is generated; a responsibility anchoring coefficient is calculated according to the report, semantic reconstruction and iterative optimization are performed on the initial metaphor, and multiple candidate consensus expressions are generated; the robustness of the responsibility distribution under each candidate expression is tested by simulating an extreme cognitive bias and a long-time-span stress situation, and an expression that is most stable and satisfies responsibility convergence and traceability conditions is selected as an output of a final consensus framework; the application outputs a consensus framework with clear responsibility anchor points and anti-interference capability.

Description

Technical Field

[0001] This invention relates to the field of natural language processing technology, specifically to a method for constructing a consensus on group risk perception based on risk metaphor. Background Technology

[0002] In complex decision-making areas involving multiple stakeholders, such as public policy making, technology ethics governance, and environmental risk assessment, building effective collective consensus on risk perception is crucial for guiding rational dialogue and achieving collaborative action. Currently, in discussions about abstract, uncertain, or long-term risks, groups often rely on metaphors for communication and understanding. These risk metaphors are not merely rhetorical tools; they profoundly shape people's cognitive framework regarding the causes of risks, attribution of responsibility, and coping strategies.

[0003] Existing technologies lack methods to identify and avoid the unintentional, systematic implantation of pre-existing frameworks that lead to "structural dilution of responsibility cognition" during the metaphorical construction of collective risk perception consensus, driven by the pursuit of superficial harmony and acceptability. Specifically, existing consensus-building methods often stop at guiding parties to accept a neutral new metaphor (such as referring to gene editing as "editing the text of life"), failing to grasp the implicit cognitive assumptions of "reversibility" and "revisionability" within the metaphorical framework itself. These assumptions naturally lead to a light regulatory path that emphasizes process management rather than fundamental prohibition in subsequent governance decisions, thus subtly transferring and dispersing the ultimate, centralized prudent responsibility that current decision-makers should bear to countless uncertain "revisers" or execution stages in the future. This invention, through quantitative analysis using "responsibility flow topology mapping," can diagnose these hidden defects of responsibility dissipation and broken backtracking paths in metaphorical frameworks before consensus is reached, and correct them at the source of cognition, ensuring that the constructed consensus has a solid anchor of ethical responsibility. Summary of the Invention

[0004] The purpose of this invention is to provide a method for constructing a collective risk perception consensus based on risk metaphor, so as to solve the problems mentioned above.

[0005] The objective of this invention can be achieved through the following technical solutions: The method for building a consensus on group risk perception based on risk metaphor includes the following steps: S1: Collect multi-source interactive texts of the group under the preset risk issues, extract metaphorical elements and semantic relationships related to risk perception and response, and form an initial metaphorical cognition set; S2: Based on the initial metaphorical cognition set, create a dynamic relationship structure that represents the responsibility association and time dimension of different risk cognition subjects. The responsibility status of the dynamic relationship structure is defined as the attribute quantity transmitted in the network. S3: Calculate the transmission path of attributes in the dynamic relationship structure, identify the dispersion and backtracking characteristics of the transmission path, and generate a quantitative analysis report on the distribution status of responsibility. S4: Based on the quantitative analysis report, semantic reconstruction and iterative optimization are performed on the core elements in the initial metaphorical cognition set to generate multiple candidate consensus metaphorical expressions with different responsibility constraints. S5: Perform structural verification on the candidate consensus metaphors, and select the expressions that enable the responsibility distribution to meet the preset convergence and traceability conditions, and output them as the final risk cognition consensus framework.

[0006] As a further aspect of the present invention: S1 specifically includes: Syntactic parsing and dependency analysis are performed on multi-source interactive texts to identify and extract core triples containing action, object and attribute modifiers, and the core triples represent specific metaphorical expressions. Using core triples as nodes, based on the co-occurrence context of core triples, syntactic dependencies, and a pre-defined domain knowledge graph, the semantic association strength between nodes is calculated, and a metaphorical semantic association network is dynamically constructed. In the metaphor semantic association network, an adversarial perturbation vector is introduced to simulate the potential cognitive bias of different risk perception positions, calculate the structural stability of the metaphor semantic association network under perturbation, and identify the fragile semantic paths in which the association strength changes beyond a preset threshold. Based on the structural stability calculation results, fragile semantic paths in the metaphor semantic association network are filtered out, and the core triples and semantic relations associated with the remaining stable paths are output as the initial metaphor cognition set.

[0007] As a further aspect of the present invention: S2 specifically includes: Each set of semantic relations in the initial set of metaphorical cognition is mapped to a directed connection, where the agent who initiates the action is the starting node and the recipient of the action or the affected object is the ending node, thus forming an initial relation network. Each directed connection in the initial relational network is assigned an initial attribute value, which is determined based on the semantic strength of the connection and the preset persistence of the semantic strength over time. In the initial relational network, based on the network topology, the process of transferring and superimposing the initial attribute quantities along the directed connections to subsequent nodes is iteratively calculated to generate an attribute quantity distribution map. Based on the attribute distribution map, key nodes for attribute convergence and edge paths for dissipation are identified, generating a dynamic relationship structure that represents responsibility association and time dimension.

[0008] As a further aspect of the present invention: the generation of the attribute quantity distribution map specifically includes: Based on the pre-defined persistence of the time dimension associated with directed connections, a decay factor that increases with time distance is calculated for each connection in the network. Based on the connection direction and attenuation factor defined by the network topology, the initial attribute value of the starting node is multiplied by the attenuation factor and then passed to the directly connected subsequent nodes. In each round of propagation calculation, when a node receives attribute values ​​from multiple predecessor nodes, a weighted summation operation is performed, and the summation result is used as the new attribute value of the corresponding node in the current round. The new attribute value is used as input, and the process of passing calculation and weighted superposition is repeated until the change of the attribute value in the initial relationship network is lower than the preset threshold. At this point, the iteration stops and the final attribute value distribution map is generated.

[0009] As a further aspect of the present invention: S3 specifically includes: Based on the attribute distribution map, extract all complete transmission paths from the start node to the end node; For each transmission path, the rate of change of attribute values ​​from the starting node to the ending node is calculated as a dispersion index of the transmission path. In the attribute distribution map, explore in the opposite direction of each transmission path, identify the connection structure that forms a closed loop, and calculate the sum of attribute quantities within the closed loop as the backtracking potential. Based on the dispersion index and backtracking potential of all transmission paths, a structured report is generated, identifying paths with high dispersion index and backtracking potential below a preset threshold as high-risk liability dilution paths.

[0010] As a further aspect of the present invention: the exploration along the opposite direction of each transmission path in the attribute quantity distribution map to identify the connection structure forming a closed loop specifically includes: Starting from any node in the attribute distribution graph, traverse in the opposite direction of the directed connection, record the nodes traversed, and form a backtracking node sequence. During the backtracking traversal, it is detected in real time whether there are duplicate nodes in the current backtracking node sequence. When a duplicate node is detected, it is determined that the path segment between the duplicate node and the first occurrence position of the duplicate node in the sequence constitutes a candidate closed loop structure. Extract all directed connections contained in the candidate closed-loop structure and summarize the attribute values ​​carried by these connections to transfer weights. The aggregated attribute weights are compared with a preset weight threshold, and candidate closed-loop structures that are not lower than the preset weight threshold are confirmed as valid responsibility backtracking closed loops.

[0011] As a further aspect of the present invention: S4 specifically includes: Based on the dispersed indicators and retrospective potential in the quantitative analysis report, a responsibility anchoring coefficient is calculated for each core semantic element to enhance responsibility constraints. Using the responsibility anchoring coefficient as a moderating parameter for semantic reconstruction, the semantic relationships of core elements in the initial metaphorical cognition set are vectorized and reorganized to generate a preliminary set of metaphorical expressions. Based on the topological characteristics of dynamic relational structures, the attribute quantity transfer process of preliminary metaphorical representations in the structure is simulated, and the generated responsibility distribution is evaluated to determine whether it conforms to preset constraints. The evaluation results are fed back to the calculation step of the responsibility anchoring coefficient. The calculation of the responsibility anchoring coefficient and semantic reconstruction are performed iteratively until multiple candidate consensus metaphor tables that meet the preset constraints are generated.

[0012] As a further aspect of the present invention: the calculation process of the responsibility anchoring coefficient is as follows: Extract the dispersion indicators of all transmission paths associated with each core semantic element in the quantitative analysis report, and calculate the arithmetic mean of all dispersion indicators. Extract the backtracking potential of all valid responsibility backtracking loops associated with the same core semantic element in the quantitative analysis report, and calculate the weighted average of all backtracking potentials; The weighted average backtracking potential energy is divided into its reciprocal to obtain the responsibility backtracking strength. The responsibility backtracking strength is then multiplied by the arithmetic average dispersion index to obtain the preliminary constraint strength value. The initial constraint strength value is amplified based on the topological centrality of the core semantic element in the dynamic relation structure. The amplified result is then normalized to obtain the responsibility anchoring coefficient.

[0013] As a further aspect of the present invention: S5 specifically includes: Reconstruct the dynamic relation structure corresponding to each candidate consensus metaphor representation, and calculate the attribute quantity distribution in the dynamic relation structure; In the dynamic relationship structure, multiple preset stress situation parameters are introduced. These stress situation parameters simulate extreme cognitive biases and long time spans, and the attribute quantity transfer process is iteratively calculated along the virtual timeline. Record the key change nodes of attribute distribution during the iterative calculation process, and calculate the statistical difference of the final responsibility distribution under different stress scenarios; The statistical dissimilarity is compared with a preset stability threshold, and the candidate consensus metaphor representation with the lowest dissimilarity that meets the preset convergence and traceability conditions is selected as the final output.

[0014] As a further aspect of the present invention: the process of iteratively calculating the attribute quantity transfer along the virtual timeline specifically includes: Based on the preset extreme cognitive bias type, the target node and directed connection for applying stress situation parameters in the dynamic relationship structure are determined; Apply numerical perturbations of the corresponding deviation type to the initial attribute quantities or transmission efficiency of the target node and the directed connection to form a structural state affected by pressure. Based on the simulation requirements spanning a long period of time, the virtual timeline is divided into multiple consecutive time windows, and a natural decay rule for the attribute quantity is set within each time window. Starting from the initial time window, the transmission process of attribute quantities along the directed connections is iteratively calculated according to the decay rule in the structural state under pressure until the entire virtual timeline is covered.

[0015] The beneficial effects of this invention are: 1. This invention transforms metaphorical elements expressed in natural language into structured networks and calculable numerical indicators (such as semantic association strength, attribute quantity, dispersion index, and backtracking potential), enabling quantitative mapping and bottleneck diagnosis of group cognitive states. This method not only automatically filters out unstable and emotional interference components in expressions (such as "fragile semantic paths"), but also accurately locates the core contradictions leading to blurred or diluted responsibility cognition (i.e., "high-risk responsibility dilution paths"). Based on this, through targeted semantic reconstruction and iterative optimization based on responsibility anchoring coefficients, it can systematically generate multiple consensus options that are strengthened in terms of responsibility constraints. This changes the inefficient traditional model of "repeated discussions and compromises," transforming the consensus-building process from an art dependent on experience and rhetoric into an analyzable, designable, and optimizable technical process.

[0016] 2. This invention introduces multi-scenario stress testing (simulating extreme cognitive biases and long time spans) in the final stage to rigorously "stress test" the stability of responsibility distribution in candidate consensus frameworks, selecting the optimal framework that maintains responsibility convergence and traceability even under the most unfavorable conditions. This ensures that the final consensus framework is not a static, ideal description, but a dynamic structure that has undergone "robust training" and can anticipate and withstand future cognitive shocks. The clear responsibility convergence nodes, backtracking paths, and anti-interference characteristics embedded in this framework can directly provide a solid and operable cognitive blueprint and decision-making basis for subsequent risk communication strategy formulation, responsibility allocation, and emergency plan design. Attached Figure Description

[0017] The invention will now be further described with reference to the accompanying drawings.

[0018] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1 As shown, this invention is a method for constructing a collective risk perception consensus based on risk metaphor, comprising the following steps: S1: Collect multi-source interactive texts of the group under the preset risk issues, extract metaphorical elements and semantic relationships related to risk perception and response, and form an initial metaphorical cognition set; S2: Based on the initial metaphorical cognition set, create a dynamic relationship structure that represents the responsibility association and time dimension of different risk cognition subjects. The responsibility status of the dynamic relationship structure is defined as the attribute quantity transmitted in the network. S3: Calculate the transmission path of attributes in the dynamic relationship structure, identify the dispersion and backtracking characteristics of the transmission path, and generate a quantitative analysis report on the distribution status of responsibility. S4: Based on the quantitative analysis report, semantic reconstruction and iterative optimization are performed on the core elements in the initial metaphorical cognition set to generate multiple candidate consensus metaphorical expressions with different responsibility constraints. S5: Perform structural verification on the candidate consensus metaphors, and select the expressions that enable the responsibility distribution to meet the preset convergence and traceability conditions, and output them as the final risk cognition consensus framework.

[0021] In S1, text data related to preset risk topics is collected from social media, news comments, forum discussions, and meeting minutes via application programming interfaces (APIs) or web crawlers. After removing irrelevant symbols and segmenting the text into sentences, a structured corpus is formed. Automatic grammatical parsing and dependency analysis are performed on each sentence. Specifically, word segmentation and syntactic analysis tools are used to identify verbs or gerunds as "actions," their directly governed nominal components as "objects," and adjectives, adverbs, or prepositional phrases modifying the action or object as "attribute modifiers." These three are combined sequentially to form a "core triple." For example, the sentence "We fight against an unknown virus" is parsed to obtain the triple (fight against, virus, unknown).

[0022] A metaphorical semantic association network is constructed using all core triples as nodes. The semantic association strength between nodes is calculated through the following steps: 1) Count the number of times two triples co-occur within the same context window and calculate their co-occurrence frequency; 2) Determine if there is a direct dependency relationship between the two triples in the original syntax, and assign a fixed weight if so; 3) Query a pre-constructed domain knowledge graph, and assign additional weights if the action or object concepts in the two triples have a direct relationship or a common superordinate concept. The three weight values ​​are then added together and divided by 3 to obtain the normalized association strength, which serves as the edge weight connecting the two nodes.

[0023] To evaluate network stability, adversarial perturbations are introduced. A random perturbation vector is generated for each node, with a random direction and a magnitude ranging from 10% to 20% of the original semantic vector magnitude of that node. This perturbation vector is superimposed on the semantic representation of the node, and the association strength of all edges is recalculated on the perturbed network. The absolute value of the change in strength of each edge before and after the perturbation is compared. If this absolute value is greater than a preset stability threshold of 0.15, the edge is marked as a "fragile semantic path". This threshold is determined through multiple random perturbation experiments, selecting a boundary value that keeps the critical connections of the network stable.

[0024] Finally, all edges marked as fragile semantic paths are removed from the original metaphorical semantic association network. If a node is no longer connected to other nodes in the network due to edge deletion, that node is also deleted. The remaining nodes and the core triples and their semantic relations in the subnetwork formed by their connections are output as the initial metaphorical cognition set.

[0025] In S2, an initial relational network is constructed based on the initial set of metaphorical cognitions. Each set of semantic relations in the set is mapped to a directed connection. Specifically, the core triplet that initiates the action in the semantic relation is taken as the starting node of the directed connection, and other core triplets that receive or are affected by the action are taken as the ending nodes. All directed connections generated according to this rule together constitute a directed graph, called the initial relational network.

[0026] Each directed connection in the initial relational network is assigned an initial attribute value. This attribute value is a non-negative real number, and its assignment is based on two factors: first, the semantic strength of the semantic relation from which the directed connection originates within the initial metaphorical cognitive set, which has already been calculated in the first step; and second, the presupposed persistence of the responsibility implied by the semantic relation over the time dimension. The presupposed persistence is pre-leveled according to domain knowledge; for example, a relation representing immediate action is assigned level 1, and a relation representing long-term consequences is assigned level 3. The initial attribute value is specifically calculated by multiplying the semantic strength value by the persistence level value.

[0027] Iterative propagation calculations of attribute quantities are performed in the initial relation network to generate an attribute quantity distribution map. The specific process is as follows: First, calculate the propagation attenuation factor. Calculate an attenuation factor for each directed connection in the network. If the preset persistence level is a value between two integers, define both the integer and fractional parts. Obtain the baseline attenuation value corresponding to the integer part, and obtain the baseline attenuation value corresponding to the next integer level. Calculate the difference between these two baseline attenuation values. Then, multiply this difference by the fractional part of the preset persistence level, and add the resulting product to the baseline attenuation value corresponding to the lower integer level. The final sum is the attenuation factor calculated for that directed connection. This factor is a positive number less than 1, and its value is related to the preset persistence level associated with the connection. The higher the preset persistence level, the closer the attenuation factor is to 1, indicating that the attribute value decays more slowly during propagation. For example, a connection with a persistence level of 1 can have an attenuation factor set to 0.5; a connection with a persistence level of 3 can have an attenuation factor set to 0.9.

[0028] Second, perform a single-round attribute transfer. Starting from all initial nodes, multiply the current attribute value of each node by the decay factor corresponding to each outgoing directed connection, and pass the result to the subsequent nodes pointed to by that connection.

[0029] Third, the node attribute values ​​are summed. When a node receives attribute values ​​from multiple predecessor nodes in the same round of computation, these incoming attribute values ​​are summed, and the sum is added to the node's original attribute values. The resulting total is used as the node's new attribute values ​​after the end of this round of computation.

[0030] Fourth, iteration and termination. The new attribute values ​​of all nodes calculated in this round are used as input for the next round of calculation, and the above transfer and superposition process is repeated. This iteration continues until the maximum change in the attribute values ​​of all nodes in the entire network between two adjacent rounds is lower than a preset convergence threshold (e.g., 0.01). At this point, the iteration stops, and the final attribute values ​​of all nodes are recorded to form an attribute value distribution map.

[0031] Finally, based on the attribute distribution map, the responsibility flow structure is analyzed. All nodes in the map are traversed, and nodes with final attribute values ​​significantly higher than the network average are identified and marked as key nodes for responsibility convergence. Simultaneously, connection paths in the network are analyzed, and paths where attribute values ​​decay to near zero after iteration are marked as edge paths of responsibility dissipation. By combining the information from key nodes and edge paths, the concentration and dissipation of attribute values ​​(i.e., responsibility states) in the network are clarified, thereby generating a dynamic relationship structure that clearly characterizes the strength of responsibility associations and the time decay effect.

[0032] In S3, all complete responsibility propagation paths are extracted based on the attribute distribution graph. Specifically, this involves identifying all starting nodes without predecessors and all ending nodes without successors from the graph. A starting node without a predecessor is selected from the attribute distribution graph as the starting point for the current search. From this node, each of its direct successor nodes is visited sequentially along all its outgoing directed connections. When visiting each successor node, the currently traversed node sequence is recorded as a temporary path. It is then determined whether the successor node is an ending node (i.e., a node without outgoing connections). If it is an ending node, the current temporary path is recorded as a complete propagation path. If it is not an ending node, this node is used as the new starting node, and the above process is repeated: continuing to visit the next successor node along its outgoing connections and updating the temporary path. When all subsequent branches originating from a certain node have been explored, the process backtracks to the previous node and attempts to explore other subsequent branches that have not yet been visited. This process continues until all branches leading to the final node from the selected starting node have been explored and all corresponding paths have been recorded. Then, the next starting node that has not yet been explored is selected, and the entire search process is repeated until all starting nodes in the graph have been traversed. In this way, all complete propagation paths existing in the graph can be systematically enumerated.

[0033] For each extracted transmission path, its responsibility dispersion index is calculated. The dispersion index quantifies the degree of attenuation of attribute quantity (i.e., responsibility intensity) from the starting point to the ending point of the path. The calculation process is as follows: obtain the final attribute quantity of the starting node in the attribute quantity distribution map, denoted as the starting value; obtain the final attribute quantity of the ending node in the map, denoted as the ending value. Then, calculate the ratio of the ending value to the starting value. Finally, subtract this ratio from 1; the result is the dispersion index for that path. This index is a value between 0 and 1; the higher the value, the more severe the dissipation of responsibility along the path.

[0034] In the attribute distribution map, a reverse exploration is performed along each transmission path to identify potential responsibility backtracking loops and calculate their backtracking potential. This process specifically includes: First, closed-loop structure identification. Starting from any node on the path, traverse in the reverse direction of the directed connections (i.e., from the end node to the start node). During the traversal, record the nodes visited in sequence, forming a backtracking node sequence. During traversal, check in real-time whether the currently visited node already exists in this sequence. If a duplicate node is found, determine that all nodes and connections between the first and second occurrences of the duplicate node constitute a candidate closed-loop structure.

[0035] Second, confirm the validity of the closed loop. Extract all directed connections contained in the candidate closed loop structure, and sum the decay factors used in the last iteration of these connections in the second step of the calculation. The sum is used as the attribute transfer weight sum of the closed loop. Compare this weight sum with a preset weight threshold (e.g., 0.7). If the weight sum is not lower than this threshold, the closed loop structure is considered to have sufficient transfer power to form a meaningful responsibility backtracking loop, thus confirming it as a valid responsibility backtracking closed loop.

[0036] Third, calculate the backtracking potential energy. For an effective responsibility backtracking loop, sum the final attribute values ​​of all nodes in the attribute distribution map. The sum is the backtracking potential energy of the loop. The level of the backtracking potential energy reflects the strength of responsibility circulating and accumulating within the loop without easily spilling out.

[0037] Finally, a structured quantitative analysis report is generated. The report summarizes the dispersion index of each transmission path and the backtracking potential of all associated valid closed loops in a list format. For each path, its dispersion index is compared to a preset high-risk dispersion threshold (e.g., 0.6), while the highest backtracking potential in its associated closed loops is compared to a preset low backtracking potential threshold (e.g., 5.0). The report explicitly identifies transmission paths with dispersion indices above the high-risk dispersion threshold and highest backtracking potential below the low backtracking potential threshold; these paths are classified as "high-risk responsibility dilution paths." This report reveals the weaknesses in the dynamic relationship structure where responsibility is easily lost and difficult to trace.

[0038] In S4, the purpose of this step is to modify the initially extracted metaphorical cognition based on the quantitative analysis report, aiming to strengthen its responsibility constraint attribute, and to form multiple feasible consensus expression candidate schemes through iterative screening.

[0039] First, a responsibility anchoring coefficient is calculated for each core semantic element in the initial metaphorical cognition set. This coefficient is a positive value and is used to quantify the degree to which the element needs to be strengthened in subsequent semantic reconstruction. The calculation of the coefficient depends on the associated data of the element in the quantitative analysis report, and the specific process is as follows: Data extraction and preprocessing: For the current core semantic element, identify all transmission paths that include it from the quantitative analysis report and extract the corresponding dispersion indicators for these paths. Simultaneously, identify all valid responsibility backtracking loops containing this element and extract the backtracking potential of these loops and the number of directed connections constituting the loops.

[0040] Calculate the average of the core indicators: Average of the dispersed indicators: Sum all extracted dispersed indicators, then divide by the total number of dispersed indicators to obtain the arithmetic mean, denoted as . Weighted average of backtracking potentials: Each extracted backtracking potential is assigned a weight equal to the number of directed connections in its closed loop. The product of each backtracking potential and its corresponding weight is calculated, all products are summed, and this sum is divided by the sum of all weights to obtain the weighted average, denoted as . .

[0041] Calculate the initial constraint strength value: Initial constraint strength value From the average of the dispersion index and the backtracking potential energy weighted average The joint decision aims to reflect the theoretical constraints on responsibility at that element. Its calculation formula is as follows: ;in, Known as the strength of responsibility retrospection, it reflects the weaker the responsibility retrospection capability and the lower the retrospection potential energy. The larger its reciprocal, the more urgent the need to enhance retrospection attributes through semantic reconstruction. This value is further correlated with the value reflecting the degree of responsibility dissipation. Multiplying them together yields the initial constraint strength.

[0042] Introducing topological centrality for correction and normalization: To reflect the differences in importance of core elements in the network structure, their topological centrality needs to be considered. Topological centrality The calculation is based on the connectivity of the node corresponding to the core semantic element within the dynamic relation structure generated in the second step. One specific calculation method is to calculate the weighted degree centrality of the node, which is the sum of the initial attribute values ​​of all directed connections (including incident and outgoing connections) to that node. Then, the initial constraint strength value is amplified, using the following formula: Finally, the amplified constraint strength values ​​obtained from calculating all core semantic elements are... Normalization is performed to ensure that the values ​​fall between 0 and 1. The normalization method involves adjusting the settling factor of an element. Divide by the largest of all elements The value is the responsibility anchoring coefficient of the core semantic element after normalization. .

[0043] Perform vectorization adjustment and reorganization of semantic relationships. Represent each core semantic element in the initial metaphorical cognitive set and its relationship with other elements as a vector in a high-dimensional semantic space. Use the responsibility anchoring coefficient. As a regulating parameter, the semantic relation vector is adjusted. Specifically, for each responsibility association that needs strengthening (e.g., strengthening the backtracking attribute based on the report), an adjustment vector pointing in the direction of "stronger responsibility constraint" is calculated in the semantic space, and then this adjustment vector is multiplied by the corresponding responsibility anchoring coefficient. This is then superimposed onto the original semantic relation vector. After performing this type of directional adjustment on all relevant semantic relations within the set, a new set of preliminary metaphorical representations containing different intensities of responsibility constraints are generated based on the adjusted vector relations.

[0044] Each generated initial metaphorical representation is simulated and evaluated. The new metaphorical representations are mapped back to the dynamic relational structure, and the transmission and distribution of responsibility (attribute quantity) in the new structural relation are simulated according to the attribute quantity transfer calculation rules established in the second step. The new dispersion index and backtracking potential generated after the simulation are calculated, and their compliance with preset constraints is evaluated. The preset constraints are usually specified as several quantifiable objectives, such as: "the dispersion index of all major transmission paths should be lower than 0.5" and "there is at least one effective responsibility backtracking loop with a backtracking potential higher than 8.0".

[0045] Perform iterative optimization. Use the simulation evaluation results as feedback input. If the generated metaphorical representation does not meet the preset constraints, fine-tune the calculation parameters of the responsibility anchoring coefficient (e.g., adjust the weights of the weighted average) based on the specific reasons for the non-compliance (e.g., whether the dispersion index is too high or the backtracking potential is too low), and then restart the entire process from calculating the responsibility anchoring coefficient to the simulation evaluation. This cycle continues until a specified number (e.g., 3 to 5) of metaphorical representations are generated, all of which meet the preset constraints but with slightly different emphases or strengths of the responsibility constraints. These representations serve as the final candidate consensus metaphorical representation output for subsequent steps to verify and select.

[0046] In S5, the dynamic relational structure corresponding to each candidate consensus metaphorical representation is reconstructed, and the attribute distribution in this structure is calculated. The specific implementation process is the same as in step two: the starting and ending nodes of directed connections are determined based on the semantic relations of the representation, forming an initial relational network; initial attribute values ​​based on semantic strength and preset persistence are assigned to the directed connections; iterative calculation of attribute values ​​is performed in the network until the attribute value change is below the convergence threshold (0.01), thereby generating a unique attribute distribution map for the candidate representation. This step establishes a baseline for subsequent stress testing.

[0047] Within the established dynamic relationship structure, multiple pre-defined stress scenario parameters are introduced to simulate the impact of extreme cognitive biases and long time spans that may occur in the real world. The transfer process of attribute quantities is then iteratively calculated along a virtual timeline. Specifically, this includes: First, identify the points of pressure application. Based on the pre-defined types of extreme cognitive biases (e.g., "overly optimistic" or "extremely panicked"), determine the target nodes and directed connections to exert influence within the dynamic relationship structure. The "overly optimistic" bias might select nodes that characterize the consequences of risk, reducing their initial attribute values; the "extremely panicked" bias might select nodes that characterize the sources of risk, increasing their initial attribute values, or select key connections, reducing their transmission efficiency (i.e., attenuation factor).

[0048] Second, numerical perturbations are applied to create a structural state under stress. For selected target nodes, their initial attribute values ​​are multiplied by a perturbation coefficient greater than or less than 1 (e.g., 1.5 for simulating "extreme panic" and 0.6 for simulating "overly optimistic"). For selected directed connections, their attenuation factors are multiplied by a coefficient less than 1 (e.g., 0.8) to simulate obstructed responsibility transmission paths.

[0049] Third, set simulation rules for long time spans. Divide the entire virtual timeline into multiple consecutive and equal-length time windows (e.g., 10 windows, each representing a time period). Set a global natural decay rate (e.g., after each time window, the current attribute value of all nodes in the network is multiplied by 0.95) to simulate the trend of responsibility fading or dispersing naturally over time.

[0050] Fourth, perform iterative calculations under stress. Starting from the first time window, using the perturbation-adjusted structural state (i.e., the adjusted initial attribute quantities and decay factors) as the starting point, calculate the attribute quantity distribution within a time window according to the transfer and superposition rules in step two. Then, multiply the calculation result by the natural decay rate (0.95) corresponding to that window. Use this result as the input attribute quantity for the next time window, and repeat the transfer, superposition, and decay process until all time windows (e.g., 10) are simulated, obtaining the final attribute quantity distribution under stress.

[0051] Record and analyze the simulation process, and calculate the statistical dissimilarity. During the iterative calculations along the virtual timeline, record key nodes where the attribute quantity distribution changes significantly. The specific criterion is: if the relative change in attribute quantity between two adjacent time windows exceeds 20%, record that node and its corresponding time window. For each applied stress scenario, obtain a final attribute quantity distribution under stress. Calculate the statistical dissimilarity between this distribution and the baseline attribute quantity distribution (without stress) obtained in the first step. The dissimilarity is calculated using the average absolute value of the differences in attribute quantities across all nodes.

[0052] The selection and output are based on the degree of difference. Multiple statistical dissimilarity values ​​calculated for each candidate consensus metaphor under different stress scenarios are compared with a preset stability threshold (e.g., 0.15). Candidate representations whose statistical dissimilarity values ​​are below this stability threshold under all stress scenarios are selected. Among these robust candidate representations, their baseline attribute distributions are further examined to see if they meet preset convergence and traceability conditions (e.g., the existence of at least one effective responsibility backtracking loop with a backtracking potential greater than 5.0, and the main path dispersion index being below 0.55). The candidate consensus metaphor representation that fully meets the conditions and has the lowest average statistical dissimilarity value is selected as the final risk perception consensus framework and output.

[0053] The working principle of this invention is as follows: Metaphorical elements are extracted from multi-source interactive texts, and a semantic network is constructed. Adversarial perturbations are introduced to filter unstable associations, forming an initial set of metaphorical cognitions. Next, this set is mapped to a dynamic relational network, quantifying responsibility as attributes that are transmitted and decayed between network nodes. A structural graph representing the flow of responsibility is generated through iterative calculation. The responsibility transmission paths in this structure are calculated, quantifying their dispersion and traceability, identifying high-risk responsibility dilution paths, and generating a quantitative analysis report. Based on this report, the responsibility anchoring coefficient of each core element is calculated, and the initial metaphors are semantically reconstructed and iteratively optimized accordingly, generating multiple candidate consensus statements that strengthen responsibility constraints. By simulating extreme cognitive biases and long-term stress scenarios, the robustness of responsibility distribution under each candidate statement is tested. The most stable statement, which maintains responsibility convergence and traceability under stress, is selected as the final risk cognition consensus framework output. This method realizes the transformation from subjective, vague risk metaphors to an objective, structured, and interference-resistant consensus framework.

[0054] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for constructing a collective risk perception consensus based on risk metaphor, characterized in that, Includes the following steps: S1: Collect multi-source interactive texts of the group under the preset risk issues, extract metaphorical elements and semantic relationships related to risk perception and response, and form an initial metaphorical cognition set; S2: Based on the initial metaphorical cognition set, a dynamic relationship structure representing the responsibility association and time dimension of different risk cognition subjects is created. The responsibility status of the dynamic relationship structure is defined as the attribute quantity transmitted in the network, specifically including: Each set of semantic relations in the initial set of metaphorical cognition is mapped to a directed connection, where the agent who initiates the action is the starting node and the recipient of the action or the affected object is the ending node, thus forming an initial relation network. Each directed connection in the initial relational network is assigned an initial attribute value, which is determined based on the semantic strength of the connection and the preset persistence of the semantic strength over time. In the initial relational network, based on the network topology, the process of transferring and superimposing the initial attribute quantities along the directed connections to subsequent nodes is iteratively calculated to generate an attribute quantity distribution map. Based on the attribute distribution map, key nodes for attribute convergence and edge paths for dissipation are identified, and a dynamic relationship structure representing responsibility association and time dimension is generated; S3: Calculate the attribute transmission paths in the dynamic relationship structure, identify the dispersion and backtracking characteristics of the transmission paths, and generate a quantitative analysis report on the responsibility distribution status, specifically including: Based on the attribute distribution map, extract all complete transmission paths from the start node to the end node; For each transmission path, the rate of change of attribute values ​​from the starting node to the ending node is calculated as a dispersion index of the transmission path. In the attribute distribution map, explore in the opposite direction of each transmission path, identify the connection structure that forms a closed loop, and calculate the sum of attribute quantities within the closed loop as the backtracking potential. Based on the dispersion index and backtracking potential of all transmission paths, a structured report is generated, identifying paths with high dispersion index and backtracking potential below a preset threshold as high-risk liability dilution paths. S4: Based on the quantitative analysis report, semantic reconstruction and iterative optimization are performed on the core elements in the initial metaphorical cognition set to generate multiple candidate consensus metaphorical expressions with different responsibility constraints. S5: Perform structural verification on the candidate consensus metaphors, and select the expressions that enable the responsibility distribution to meet the preset convergence and traceability conditions, and output them as the final risk cognition consensus framework.

2. The method for constructing a consensus on group risk perception based on risk metaphor as described in claim 1, characterized in that, S1 specifically includes: Syntactic parsing and dependency analysis are performed on multi-source interactive texts to identify and extract core triples containing action, object and attribute modifiers, and the core triples represent specific metaphorical expressions. Using core triples as nodes, based on the co-occurrence context of core triples, syntactic dependencies, and a pre-defined domain knowledge graph, the semantic association strength between nodes is calculated, and a metaphorical semantic association network is dynamically constructed. In the metaphor semantic association network, an adversarial perturbation vector is introduced to simulate the potential cognitive bias of different risk perception positions, calculate the structural stability of the metaphor semantic association network under perturbation, and identify the fragile semantic paths in which the association strength changes beyond a preset threshold. Based on the structural stability calculation results, fragile semantic paths in the metaphor semantic association network are filtered out, and the core triples and semantic relations associated with the remaining stable paths are output as the initial metaphor cognition set.

3. The method for constructing a consensus on group risk perception based on risk metaphor as described in claim 1, characterized in that, The generated attribute distribution map specifically includes: Based on the pre-defined persistence of the time dimension associated with directed connections, a decay factor that increases with time distance is calculated for each connection in the network. Based on the connection direction and attenuation factor defined in the network topology, the initial attribute value of the starting node is multiplied by the attenuation factor and then passed to the directly connected subsequent nodes. In each round of propagation calculation, when a node receives attribute values ​​from multiple predecessor nodes, a weighted summation operation is performed, and the summation result is used as the new attribute value of the corresponding node in the current round. The new attribute value is used as input, and the process of passing calculation and weighted superposition is repeated until the change of the attribute value in the initial relationship network is lower than the preset threshold. At this point, the iteration stops and the final attribute value distribution map is generated.

4. The method for constructing a collective risk perception consensus based on risk metaphor as described in claim 1, characterized in that, The process of exploring the attribute distribution map in the opposite direction of each transmission path to identify the connection structures that form closed loops specifically includes: Starting from any node in the attribute distribution graph, traverse in the opposite direction of the directed connection, record the nodes traversed, and form a backtracking node sequence. During the backtracking traversal, it is detected in real time whether there are duplicate nodes in the current backtracking node sequence. When a duplicate node is detected, it is determined that the path segment between the duplicate node and the first occurrence position of the duplicate node in the sequence constitutes a candidate closed loop structure. Extract all directed connections contained in the candidate closed-loop structure and summarize the attribute values ​​carried by these connections to transfer weights. The aggregated attribute weights are compared with a preset weight threshold, and candidate closed-loop structures that are not lower than the preset weight threshold are confirmed as valid responsibility backtracking closed loops.

5. The method for constructing a collective risk perception consensus based on risk metaphor according to claim 1, characterized in that, S4 specifically includes: Based on the dispersed indicators and retrospective potential in the quantitative analysis report, a responsibility anchoring coefficient is calculated for each core semantic element to enhance responsibility constraints. Using the responsibility anchoring coefficient as a moderating parameter for semantic reconstruction, the semantic relationships of core elements in the initial metaphorical cognition set are vectorized and reorganized to generate a preliminary set of metaphorical expressions. Based on the topological characteristics of dynamic relational structures, the attribute quantity transfer process of preliminary metaphorical representations in the structure is simulated, and the generated responsibility distribution is evaluated to determine whether it conforms to preset constraints. The evaluation results are fed back to the calculation step of the responsibility anchoring coefficient. The calculation of the responsibility anchoring coefficient and semantic reconstruction are performed iteratively until multiple candidate consensus metaphorical representations that meet the preset constraints are generated.

6. The method for constructing a collective risk perception consensus based on risk metaphor as described in claim 5, characterized in that, The calculation process for the responsibility anchoring coefficient is as follows: Extract the dispersion indicators of all transmission paths associated with each core semantic element in the quantitative analysis report, and calculate the arithmetic mean of all dispersion indicators. Extract the backtracking potential of all valid responsibility backtracking loops associated with the same core semantic element in the quantitative analysis report, and calculate the weighted average of all backtracking potentials; The weighted average backtracking potential energy is divided into its reciprocal to obtain the responsibility backtracking strength. The responsibility backtracking strength is then multiplied by the arithmetic average dispersion index to obtain the preliminary constraint strength value. The initial constraint strength value is amplified based on the topological centrality of the core semantic element in the dynamic relation structure. The amplified result is then normalized to obtain the responsibility anchoring coefficient.

7. The method for constructing a collective risk perception consensus based on risk metaphor according to claim 1, characterized in that, S5 specifically includes: Reconstruct the dynamic relation structure corresponding to each candidate consensus metaphor representation, and calculate the attribute quantity distribution in the dynamic relation structure; In the dynamic relationship structure, multiple preset stress situation parameters are introduced. These stress situation parameters simulate extreme cognitive biases and long time spans, and the attribute quantity transfer process is iteratively calculated along the virtual timeline. Record the key change nodes of attribute distribution during the iterative calculation process, and calculate the statistical difference of the final responsibility distribution under different stress scenarios; The statistical dissimilarity is compared with a preset stability threshold, and the candidate consensus metaphor representation with the lowest dissimilarity that meets the preset convergence and traceability conditions is selected as the final output.

8. The method for constructing a collective risk perception consensus based on risk metaphor according to claim 7, characterized in that, The process of iteratively calculating attribute quantity transfer along the virtual timeline specifically includes: Based on the preset extreme cognitive bias type, the target node and directed connection for applying stress situation parameters in the dynamic relationship structure are determined; Apply numerical perturbations of the corresponding deviation type to the initial attribute quantities or transmission efficiency of the target node and the directed connection to form a structural state affected by pressure. Based on the simulation requirements spanning a long period of time, the virtual timeline is divided into multiple consecutive time windows, and a natural decay rule for the attribute quantity is set within each time window. Starting from the initial time window, the transmission process of attribute quantities along the directed connections is iteratively calculated according to the decay rule in the structural state under pressure until the entire virtual timeline is covered.

Images