A Method for Generating Medical Ethics Analysis Reports Based on Multi-Model Game Theory

Abstract

This invention discloses a method for generating medical ethics analysis reports based on multi-model game theory, comprising: analyzing medical ethics cases and constructing a structured element set and a role-based model configuration set; implementing multi-model game theory, dynamically evolving the argument strength through an excited-state decay mechanism to obtain an evolutionary argument strength matrix, game trajectory data, and resonance suppression records; analyzing the evolutionary argument strength matrix and game trajectory data to mine the fusion of viewpoints and crystallize structural disagreements, obtaining a fused viewpoint set, a crystallized disagreement set, and an ethical tension map; and integrating these to generate a medical ethics analysis report. This invention improves the dynamic authenticity of the medical ethics analysis process and the objectivity of the results.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence and data processing, and in particular to a method for generating medical ethics analysis reports based on multi-model game theory. Background Technology

[0002] With the emergence of cutting-edge technologies such as AI-assisted diagnosis and treatment, ethical dilemmas in clinical practice are becoming increasingly complex. These issues often involve multiple stakeholders, including patients, their families, medical staff, hospital administrators, and even the general public. Based on different views of life, values, and stances, these stakeholders are prone to differing opinions and even disagreements. Traditional ethical consultation models primarily rely on expert committees for discussion, but this model also faces several challenges in practice, such as relatively limited efficiency, significant resource investment, the potential influence of varying expert opinions on the weighting of different viewpoints, and the need for greater transparency in the decision-making process. Therefore, developing an intelligent system capable of objectively, comprehensively, and efficiently simulating multi-party debates and generating in-depth analysis reports is of significant theoretical and practical importance for improving the quality and efficiency of medical decision-making, supporting ethical education, and reducing ethical disputes.

[0003] Currently, research on intelligent medical ethics analysis has made some progress, forming various technical approaches. Early explorations mainly focused on rule-based ethical reasoning systems. These systems formalize ethical principles (such as the four principles of the Belmont Report) into logical rules, performing deductive reasoning and judgment on input cases. Meanwhile, case-based reasoning (CBR) systems have also been applied in this field, providing reference suggestions for current dilemmas by retrieving and matching similar historical cases. With the development of machine learning technology, some studies have begun to adopt supervised learning methods, training classifiers on labeled ethical case datasets to identify the type of ethical issues or assess the risk level of new cases. In recent years, the application of Large Language Models (LLM) and Multi-Agent Systems (MAS) has become a new research hotspot. Researchers are attempting to simulate ethical discussion processes by using sophisticated prompt engineering to allow a single LLM to play different roles in dialogue, or by constructing multiple agents representing different positions (such as doctors, patients, and lawyers) to interact. These methods have, to some extent, automated ethical analysis, enabling them to handle partially structured ethical scenarios and generate preliminary analysis results.

[0004] However, existing technical solutions suffer from two fundamental technical flaws in simulating the complex and dynamic ethical debates of the real world, limiting the dynamic realism of their analytical results and their value for decision-making. These two core problems are the lack of argument fatigue and the uncontrolled resonant catastrophe. This obscures more valuable and novel viewpoints, hindering the formation of genuine consensus. Summary of the Invention

[0005] The purpose of this invention is to provide a method for generating medical ethics analysis reports based on multi-model game theory, so as to solve the above-mentioned problems existing in the prior art.

[0006] The technical solution, a method for generating medical ethics analysis reports based on multi-model game theory, includes:

[0007] Analyze original medical ethics case texts to construct a structured set of ethical elements and a set of role-based model configurations;

[0008] A multi-model game was played on the set of structured ethical elements and the set of role-based model configurations. The strength of the argument was dynamically evolved through the excitation-state decay mechanism, and the evolutionary argument strength matrix, game trajectory data and resonance suppression records were obtained.

[0009] By analyzing the evolutionary argument intensity matrix and game trajectory data, we can explore the structural divergences of viewpoint fusion and crystallization, and obtain the fused viewpoint set, the crystallized divergence set, and the ethical tension map.

[0010] By integrating and fusing viewpoint sets, crystallized divergence sets, ethical tension maps, and resonance inhibition records, a medical ethics analysis report is generated.

[0011] Beneficial effects: This invention effectively prevents a single viewpoint from dominating the entire discourse in medical ethics discussions simply by repeating itself, ensuring the dynamic fairness of the game process; at the same time, it also maintains the macro-stability of the entire game system and the reliability of the analysis results. Attached Figure Description

[0012] Figure 1 A flowchart illustrating the steps of a method for generating a medical ethics analysis report based on multi-model game theory, provided in this application embodiment.

[0013] Figure 2 A flowchart illustrating the steps for obtaining the evolutionary argument intensity matrix, game trajectory data, and resonance suppression records provided in this application embodiment.

[0014] Figure 3 A flowchart illustrating the steps of the adaptive damping mechanism for suppressing resonance amplification provided in this application embodiment.

[0015] Figure 4 A flowchart illustrating the steps for obtaining a fused viewpoint set as provided in this application embodiment. Detailed Implementation

[0016] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0017] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product, or device.

[0018] The research found that existing systems generally treat the persuasiveness of each argument or viewpoint as a static or simply cumulative attribute, ignoring the argument fatigue effect. In real debates, even the most powerful argument will naturally lose its marginal persuasiveness after being repeatedly and frequently presented, leading to audience fatigue. However, existing methods cannot model this dynamic decay process, resulting in some arguments unreasonably accumulating influence simply through repeated mention. This distorts the true dynamics of the debate and obscures other potentially more valuable novel viewpoints. Furthermore, current multi-agent game systems lack effective mechanisms for identifying and controlling argument resonance. When mutually supporting arguments in the network form circular references (e.g., A supports B, B supports C, and C in turn supports A), their strength is abnormally amplified, creating resonance that allows these arguments to quickly dominate, regardless of their internal logical soundness. This uncontrolled resonance hinders the formation of genuine consensus and may lead to a biased conclusion driven by the echo chamber effect.

[0019] like Figure 1 As shown, a method for generating medical ethics analysis reports based on multi-model game theory is proposed, including the following steps:

[0020] We analyze original medical ethics case texts and construct a structured set of ethical elements and a set of role-based model configurations.

[0021] In this embodiment, this step aims to transform unstructured natural language case text into structured data that can be processed by computers, laying the foundation for subsequent quantitative game analysis. Specifically, this is achieved through two sub-steps. First, medical ethics elements are extracted, i.e., the original medical ethics case text is read, such as a case description about whether a terminally ill patient should continue to use a ventilator. Using a rule-based medical entity recognition method, basic patient information (such as age 85, advanced cancer, and confusion), treatment options (such as continuing invasive ventilation, weaning from the ventilator, and switching to palliative care) and the position statements of various stakeholders are extracted. Second, ethical dimension structured mapping is performed, reading the list of elements extracted in the previous step and performing semantic matching with a pre-built medical ethics ontology. This ontology is hierarchical; for example, the top layer consists of four principles: autonomy, benefit, harmlessness, and fairness, while the middle layer is refined into sub-principles such as informed consent and decision-making ability. The element that the patient has expressed an unwillingness to endure unnecessary suffering is mapped to the personal will expression item under the principle of autonomy, and the association strength weight is calculated. Simultaneously, it automatically identifies and labels conflicting relationships between elements, such as when the children's demand for all-out rescue conflicts with the patient's own wishes. The resulting structured set of ethical elements contains all key information and its ethical mappings and interrelationships; the role-based model configuration set assigns differentiated ethical reasoning frameworks to each subsequent AI model, such as deontological models.

[0022] A multi-model game was conducted on the set of structured ethical elements and the set of role-based model configurations. The strength of the argument was dynamically evolved through the excitation-state decay mechanism, and the evolutionary argument strength matrix, game trajectory data and resonance suppression records were obtained.

[0023] This embodiment simulates a dynamic debate among multiple parties regarding an ethical dilemma. It is not a static display of viewpoints, but rather a dynamic evolution incorporating time and interaction dimensions. Specifically, each role-based AI model first generates initial argumentative viewpoints based on its value function and structured ethical elements; for example, a deontological model might generate that a doctor has an obligation to sustain life as much as possible. Each argument is assigned an initial strength. In multiple rounds of game theory, the strength of an argument dynamically changes as it is cited or refuted by other models. An excited-state decay mechanism is introduced to simulate the fatigue effect of arguments, i.e., the persuasiveness of an argument diminishes over time when it is repeatedly used. Simultaneously, a coupling network facilitates the flow of strength among semantically related viewpoints, and an adaptive damping mechanism suppresses resonance caused by circular citations of certain viewpoints. The dynamic data of the entire process is fully recorded, forming an evolutionary argument strength matrix and other results, providing a basis for subsequent analysis.

[0024] By analyzing the evolutionary argument intensity matrix and game trajectory data, we can uncover structural disagreements that are fused and crystallized, and obtain a set of fused viewpoints, a set of crystallized disagreements, and an ethical tension map.

[0025] In this embodiment, valuable decision-making reference information is extracted from the complex game process. Specifically, by performing spectral analysis on the game trajectory data, it is possible to identify whether there are resonance frequencies among different viewpoints. If phase synchronization exists, it indicates that these viewpoints may have deep-seated common value demands, which can be mined and categorized into integrated viewpoints or consensus. Simultaneously, for argument pairs whose intensity remains consistently antagonistic in the later stages of the game, they are marked as crystallized structural disagreements, and their irreconcilability index is calculated. Finally, all viewpoints and disagreements are mapped onto a two-dimensional ethical space. By calculating the local viewpoint density and disagreement intensity, a visualized ethical tension map is generated, intuitively displaying the core contradictory areas in the case.

[0026] By integrating and fusing viewpoint sets, crystallized divergence sets, ethical tension maps, and resonance inhibition records, a medical ethics analysis report is generated.

[0027] In this embodiment, the aforementioned analysis results are presented to the user in a structured and highly readable manner. Specifically, a hierarchical report template engine transforms the fused viewpoint set into a consensus recommendation section of the report, the crystallized divergence set into a risk warning section, and the ethical tension graph serves as a decision support visualization chart. In particular, resonance inhibition records are used to generate a credibility assessment of the argumentation process, revealing to the user which arguments exhibit unstable resonance behavior in the game, thus enhancing the report's transparency and credibility. The final output is a complete, multi-dimensional, and in-depth medical ethics analysis report.

[0028] like Figure 2 As shown, according to one aspect of this application, an evolutionary argument intensity matrix, game trajectory data, and resonance suppression records are obtained, including:

[0029] Each role-based model generates an initial argument based on a structured set of ethical elements, and assigns an initial energy level to each argument through a completeness score, thereby constructing an initial argument viewpoint library and an argument association matrix that represents the semantic relationship between viewpoints.

[0030] In this embodiment, to objectively assign initial energy levels to each argument and avoid the shortcomings of subjective setting or equalization of argument strength in traditional methods, a three-dimensional scoring method combined with Softmax normalization is proposed. Specifically, for each generated structured argument text, a quantitative score is performed in the [0, 1] interval from three dimensions: the first being the clarity of premises (P... score This is used to measure whether the premises of an argument are clear and specific, for example, through formula P. score = 0.4 × normalized value of premise quantity + 0.6 × premise specificity (based on entity density) is used for calculation; the second is the logicality of reasoning (L scoreThe third aspect is the relevance of the conclusion (R). This is assessed through dependency parsing to evaluate the coherence of the reasoning steps from premise to conclusion. score This is used to calculate the semantic closeness between the conclusion and the premise. Based on this, the geometric mean method is used to calculate the overall quality score Q. total =(P score ×L score ×R score ) 1 / 3 Using geometric mean ensures that a low score in any dimension will significantly lower the overall score, preventing high-scoring arguments with major logical or premise-related flaws. This is based on the importance of the role presenting the argument (e.g., the weight w of the attending physician's role). role The quality score can be adjusted by setting the weight to 1.2 (the weight of the patient's family member role can be set to 1.0): Q adjusted =Q total ×w role To ensure the determinism of the total system energy, a softmax function is used for normalization, and the initial energy level E of each argument is calculated. 0_i = E total × exp(Q adjusted_i ) / Σ j exp(Q adjusted_j ), where E total This is the preset total system energy, for example, 10.0. This ensures the objectivity and fairness of the starting point of the game.

[0031] In multi-round game interaction, the excited-state decay model, coupled oscillator network and adaptive damping mechanism are integrated to dynamically evolve the argument strength in the initial argument viewpoint library. The excited-state decay model simulates the argument fatigue effect based on the citation history and time of the argument, the coupled oscillator network realizes the strength redistribution among related viewpoints based on the argument correlation matrix, and the adaptive damping mechanism suppresses the resonance amplification caused by circular references.

[0032] In this embodiment, this step is the dynamic core of the game process. The excited-state decay model, coupled oscillator network, and adaptive damping mechanism are treated as a whole and described by a unified differential equation, which evaluates the strength E of all arguments within each time step dt. i Perform iterative updates. The simplified unified equation form is dE. i / dt = [attenuation term] + [coupling term]. The attenuation term encompasses both the basic attenuation of the excited-state attenuation model and the effect of adaptive damping; the coupling term represents the effect of the coupled oscillator network, i.e., the mutual influence between arguments. By integrating these three mechanisms, a highly realistic debate ecosystem can be simulated. Excited-state attenuation simulates the lifecycle of individual arguments, overcoming the deficiency of existing techniques that treat argument strength as a static value; the coupled oscillator network simulates the mutual influence and clustering phenomenon between viewpoints; and adaptive damping acts as a gatekeeper of rationality, preventing the discussion from reaching a stalemate due to resonance caused by circular references.

[0033] The final result of the dynamic evolution is recorded and output as an evolutionary argument strength matrix, game trajectory data, and resonance suppression record.

[0034] In this embodiment, this step is the data accumulation stage of the game process. The evolutionary argument strength matrix is ​​a T×N matrix (T is the total number of time steps, N is the total number of arguments), recording the strength value of each argument at each time point. The game trajectory data is a richer record, also including interaction events such as citations, rebuttals, and support between arguments. The resonance suppression record is a dedicated log that records all damping events triggered during the game process, including the argument where resonance occurred, its time, strength, and the applied damping value. This output data is the direct input for subsequent analysis and report generation steps. For example, the resonance suppression record can be directly used to generate the argument process credibility assessment section of the report, a unique output not available in existing technologies.

[0035] According to one aspect of this application, the excited-state decay model simulates and demonstrates the fatigue effect in the following manner:

[0036] For any argument in the initial argument base, its argument strength E(n,t) = E0 × (1 + α×n) ×exp(-γ(n)×t) + E base .

[0037] In this embodiment, the formula is used to model the evolution of the strength of any argument over time and the number of citations. Here, E(n, t) is the real-time strength of the argument; E0 is the initial energy level of the argument; n is the cumulative number of citations of the argument during the game process; t is the evolution time since the last citation; γ(n) is the nonlinear decay rate adaptively adjusted with the number of citations n; α is a preset excitation gain coefficient; and E... baseThe preset ground state energy level is used to ensure that the argument strength does not completely disappear; for example, it can be set to 0.1×E0. Preferably, the excitation gain coefficient α is not a fixed value, but is calculated using the formula α = 0.15×exp(-0.1×n). This can simulate the real debate phenomenon of novelty preference or diminishing marginal utility: the first and first few citations of an argument bring a significant boost to its persuasiveness (excitation effect), but as the number of citations n increases, the exponential term causes the gain to decrease rapidly, solving the technical problem that existing methods cannot model the novelty preference of arguments.

[0038] According to one aspect of this application, the nonlinear decay rate γ(n) is adaptively adjusted in the following manner:

[0039] The basic attenuation rate γ0 is determined based on the argument type preset in the initial argument viewpoint library.

[0040] Calculate the citation frequency f of the argument from the citation history pattern data of the argument. cite Standard deviation σ of the reference interval distribution interval ;

[0041] The formula γ(n) = γ0×(1 +β×n) 2 ) × (1 + 0.5×log(1+f cite )) × exp(σ interval / σ max The nonlinear attenuation rate γ(n) is calculated; where β is a preset coefficient controlling the influence of the number of citations on the intensity, and σ max This is the standard deviation of the maximum citation interval used for normalization.

[0042] Specifically, to achieve personalized simulation of different argument attenuation rates, this adaptive adjustment process includes: determining the base attenuation rate γ0 based on the nature of the argument; for example, γ0 for principled arguments (such as respect for life) can be set to 0.1, while γ0 for emotional arguments (such as those unacceptable to family members) can be set to 0.2. Calculating dynamic factors related to citation history. 1+β×n 2 The term (where β can be set to 0.05) indicates that the more times an argument is cited, the faster its decay accelerates. 1 + 0.5×log(1+f cite The term indicates that excessively high citation frequency will also accelerate decay. exp(σ) interval / σ max The term ) indicates that if the time intervals at which an argument is cited are highly irregular (i.e., σ), then... interval A large value indicates that its occurrence may be random or lack logical coherence, and its decay should be accelerated accordingly. A unique decay curve was customized for each argument, enhancing the dynamic realism of the game process.

[0043] This embodiment successfully solves the problem of static argument strength by introducing an argument strength formula, bringing unprecedented dynamic realism to the ethical analysis game process. Specifically, it is achieved by using two dynamically changing data points during the game process—the cumulative number of times an argument has been cited and the time since its last citation—as core input parameters, and outputting a dynamic argument strength. In the context of medical ethics analysis, this means that an argument representing the emotional demands of family members may have high strength initially, but if this argument is repeatedly raised in subsequent discussions (n ​​increases) without new information, its strength will naturally decrease due to the decay term exp(-γ(n)×t). This accurately simulates the argument fatigue effect, preventing a few viewpoints from dominating the discussion through simple repetition, ensuring the fairness and objectivity of the game process, and enabling the final analysis report to more realistically reflect the comprehensive influence of all viewpoints after dynamic interaction, rather than a simple summation of static statements.

[0044] According to one aspect of this application, the coupled oscillator network achieves the redistribution of intensity among associated viewpoints in the following manner:

[0045] For any argument i in the initial argument base, the intensity change ΔE caused by coupling effect i The interaction effects of all other arguments j are summed to calculate ΔE. i = Σ j K ij × (E j - E i ) × Φ(semantic distance (i, j)); where E i With E j K represents the current argument strength of arguments i and j, respectively. ij To demonstrate the coupling strength derived from the correlation matrix, semantic distance (i, j) represents the semantic distance between i and j, and Φ is a preset function that maps this semantic distance to coupling weights.

[0046] In this embodiment, the formula describes the intensity change ΔE of any argument i. i How this is determined by its interaction with all other arguments j in the network. Where (E) j -E i The term Φ represents the strength potential difference between argument j and argument i. When there is a coupling relationship between the two, the strength tends to flow from the stronger argument to the weaker argument. The function Φ(semantic) distance (i, j)) is a mapping function used to map semantic distance to coupling weights. Preferably, this function can be a Gaussian function Φ(d) = exp(-d 2 / 2σ2 The form of ) (where σ can be set to 0.3) has the effect that the closer the semantic distance, the larger the weight and the stronger the coupling effect.

[0047] According to one aspect of this application, semantic distance distance The calculation methods for (i, j) include:

[0048] On a pre-constructed medical ethics concept graph, the shortest path on the graph between the core concepts of argument i and j is calculated to obtain the concept distance component;

[0049] By analyzing the text content of arguments i and j, we extract the stance vector and the core object of the argument, and calculate the stance similarity component and the overlap component of the argument object.

[0050] The semantic distance is comprehensively assessed by weighting and summing the conceptual distance component, the positional similarity component, and the overlap component of the argumentative objects. distance (i, j).

[0051] To address the overly abstract nature of existing semantic distance calculation methods, this embodiment provides a concrete and reproducible method. For example, it calculates concept distance on a pre-constructed medical ethics concept graph. This graph can be constructed as a three-layer ontology structure: the top layer comprises four ethical principles (autonomy, benefit, harmlessness, and fairness); the middle layer is refined into 12 sub-principles; and the bottom layer expands to 48 specific normative items. The distance between two core concepts of an argument on this graph is calculated using a graph shortest path algorithm, yielding the concept distance component d. concept The similarity score s was calculated through text analysis. stance (e.g., based on vector representations of support / opposition / neutrality) and the degree of overlap between the objects of discussion. object The comprehensive semantic distance is calculated by weighted summation, for example, semantic... distance (i, j) = w1×d concept + w2×(1-s stance ) + w3×(1-o object The weights can be set to w1=0.5, w2=0.3, and w3=0.2.

[0052] To demonstrate the adaptive evolution of the network topology, in a preferred embodiment, the coupling strength K... ij Instead of being based on a static value set at an initial semantic distance, it is dynamically updated using an online learning rule. Specifically, the coupling strength can be updated according to the following rule: K ij (t+1) = K ij (t) +η×(f co ×Iinteract - K ij (t)). Where η is the learning rate (e.g., 0.01), f co To demonstrate the co-occurrence frequency of i and j in the game history, I interact The coupling strength is their historical interaction strength. This mechanism enables the coupling strength to reflect the actual interaction history of arguments in the game, achieving a qualitative change from a static network to a dynamic evolutionary network.

[0053] like Figure 3 As shown, according to one aspect of this application, the adaptive damping mechanism suppresses resonant amplification in the following way:

[0054] Real-time monitoring of strongly coupled loops within the argumentation network, and calculation of an index reflecting the current loop strength for each argumentation participating in the loop;

[0055] When the cyclic strength index of any argument exceeds a preset dynamic threshold, the argument is activated and adaptive damping is applied.

[0056] Adaptive damping, as an additional attenuation term, is integrated into the intensity evolution calculation of this argument to suppress the aberrant amplification of its intensity; the triggering and effect of the suppression process are recorded in the resonance suppression record.

[0057] Specifically, the Tarjan algorithm is used to detect all strongly connected components in the network, and a depth-first search is further used to find all simple cycles. For each argument i, its cycle strength index C is calculated. i (t), which can be defined as the average of the rate of energy change in all loops involved in this argument: C i (t) = Σ j∈Loop(i) |dE j / dt| / |Loop(i)|. When C i (t) Exceeds the dynamically set threshold C threshold Damping is activated when calculated (e.g., based on historical data mean and standard deviation). The activated damping D... i (t) is added as an additional attenuation term to the evolution differential equation of the argument strength, for example, dE i / dt = (-γ i -D i (t))×E i +[coupling term], where Loop(i) is the set of all simple loops containing argument i, γ i To demonstrate the natural decay coefficient of i.

[0058] According to one aspect of this application, the adaptive damping is calculated as follows:

[0059] Through formula D i(t) = D0×[1 + tanh(μ×(C i (t) - C threshold The adaptive damping D was calculated. i (t);

[0060] Where D0 is the preset base damping coefficient, and μ is the sensitivity parameter controlling the smoothness of damping activation. This formula utilizes the properties of the hyperbolic tangent function tanh to ensure that when the cyclic strength index C... i (t) approaches and exceeds the dynamic threshold C threshold When, adaptive damping D i The value of (t) increases smoothly.

[0061] In this embodiment, to achieve smooth and stable suppression of resonance, the damping value was not calculated using a simple step function. Specifically, the basic damping coefficient D0 can be set to 0.2, and the sensitivity parameter μ can be set to 5. The range of the hyperbolic tangent function tanh is (-1, 1), and when the cyclic intensity C... i (t) is much smaller than the threshold C threshold When the tanh term is close to -1, D i When (t) approaches 0, damping has no effect; when C... i When (t) is much greater than the threshold, the tanh term approaches 1, and D i (t) approaches 2×D0, achieving the maximum damping effect; while in C i (t) The region close to the threshold, D i The value of (t) will smoothly increase from 0. This smooth transition avoids system oscillations that may be caused by damping abrupt changes, thus ensuring the macroscopic stability of the game.

[0062] This embodiment effectively solves the resonance problem by real-time monitoring of strongly coupled loops in the argumentation network and applying adaptive damping, thereby improving the stability of the game process and the reliability of the analysis results. A closed loop of monitoring-computation-suppression is formed: the system identifies circular reference structures between arguments through a loop detection algorithm; it uses game data such as the energy change rate of arguments participating in the loop as input to calculate a quantified cycle strength index; and it obtains the damping value through an adaptive damping calculation formula, integrating it as negative feedback into the evolution calculation of argument strength. In the context of medical ethics decision-making, this means that when two or more models form an echo chamber, and their viewpoints (such as prioritizing the efficiency of medical resource utilization) mutually support each other, leading to an abnormal surge in strength, this mechanism can automatically intervene and cool down the situation. This ensures that the game is not influenced by a few collaborative, highly interconnected viewpoints, making the final consensus more representative and the generated risk warnings more reliable.

[0063] like Figure 4As shown, according to one aspect of this application, a fusion viewpoint set is obtained, including:

[0064] Cross-spectral analysis was performed on the time series of interactive argument intensity in game trajectory data. By calculating the phase lock value, arguments that undergo phase synchronization at a specific resonance frequency were identified.

[0065] By tracing and merging the arguments that are in phase synchronization, and extracting common deep-seated value claims through semantic clustering, a set of integrated viewpoints is constructed.

[0066] In this embodiment, to achieve accurate analysis of the time series of cross-argument strength, preprocessing of the time series data is required. Specifically, this preprocessing includes: outlier detection and correction (e.g., using the 3σ criterion), applying a Savitzky-Golay filter (e.g., window length can be set to 21, polynomial order can be set to 3) to smooth noise, performing trend removal (e.g., subtracting first-order polynomial fitting), and segmenting using a Hamming window (e.g., window length can be set to 256, overlap rate 50%). After preprocessing, the cross-power spectral density S is calculated for the argument pairs (i, j) within each window. ij (f) =F[E i ]×F*[E j ], where F represents the Fourier transform. Phase synchronization is then detected by calculating the phase lock value (PLV), which is formulated as PLV(f) = |<exp(i×ΔΦ ij (f, t))> t |, where <> represents time averaging. When PLV(f) exceeds a preset threshold (e.g., 0.8) at a certain frequency f, it is considered that arguments i and j have achieved phase synchronization at that frequency.

[0067] Preferably, to eliminate false synchronization caused by random noise and increase the scientific rigor and reliability of consensus point identification, the method further includes calculating the p-value of each resonance frequency through a permutation test, retaining only resonance frequencies that are statistically significant (e.g., p < 0.01). For resonance frequency points that pass the test, the semantic content that triggered the resonance is traced, and related arguments are merged through semantic similarity clustering. Their common underlying value appeals are extracted (e.g., reducing patient suffering may be the common basis of multiple seemingly opposing viewpoints), thereby constructing a fused viewpoint set.

[0068] According to one aspect of this application, while identifying convergent viewpoints, a method for identifying and handling antagonistic relationships between viewpoints is also provided, specifically, obtaining a crystallized divergence set, including:

[0069] By evaluating the stability of the evolutionary argument strength matrix, we can identify argument pairs that maintain a stable adversarial state in the later stages of the game.

[0070] For each identified stable adversarial state, the root cause of the divergence is traced, and a multidimensional crystallinity index that quantifies the stability of the divergence is calculated. A crystallized divergence set containing the argument pair and the corresponding multidimensional crystallinity index is constructed.

[0071] Specifically, in order to perform stability assessment, the intensity difference time series ΔE is calculated for each pair of arguments (i, j). ij (t) = |E i (t) - E j (t)|. The local coefficient of variation (CV) of the discrepancy sequence is calculated using a sliding window. When the duration of a CV less than a preset threshold (e.g., 0.1) exceeds 60% of the total game duration, the argument pair is marked as a stable adversarial pair. When tracing the root causes of disagreements, preferably, after tracing using argument mining techniques, the root causes are automatically classified as: value premise disagreement, factual perception disagreement, or reasoning path disagreement. This provides deeper insights for ethical analysis and outputs results with clear diagnostic significance.

[0072] Furthermore, the calculation method of the multidimensional crystallinity index includes: for each stable adversarial argument pair, calculating the quantitative values ​​of four dimensions, including: temporal stability, used to quantify the degree of variation of intensity differences; intensity symmetry, used to quantify the degree of balance of intensity values; evolutionary depth, used to quantify the root depth of the disagreement; and semantic distance, used to quantify the degree of alienation of viewpoints; and obtaining the multidimensional crystallinity index by taking a weighted geometric average of the quantitative values ​​of the four dimensions.

[0073] In this embodiment, the quantization values ​​of the four dimensions are calculated as follows: Time stability dimension C time = 1-CV avg (where CV) avg (mean coefficient of variation); intensity symmetry dimension C sym = 1 - |E i - E j | / (E i + E j Evolutionary depth dimension C depth = depth / max depth (where depth is the depth of the branching tree, max) depth (the maximum depth of the branching tree); semantic distance dimension C semantic = semantic distance / max distance (where max) distance (For maximum semantic distance). The overall crystallinity C is calculated using a weighted geometric mean. total = (C timew1 × C sym w2 × C depth w3 × C semantic w4 ) (1 / Σw) The weight w can be set to [0.3, 0.2, 0.3, 0.2].

[0074] Having obtained a convergence of perspectives and crystallization of divergences, according to one aspect of this application, a method for generating an ethical tension map is provided. Specifically, the steps for obtaining the ethical tension map include:

[0075] Ethical feature vectors are extracted from the viewpoints and disagreements in the fusion viewpoint set and the crystallized disagreement set. Principal component analysis is used to determine the coordinate axes of the ethical space, and each viewpoint and disagreement is mapped to this space to obtain its coordinates.

[0076] In this ethical space, the local viewpoint density ρ and the local divergence intensity D originating from the crystallization divergence set are calculated based on each coordinate.

[0077] The local tension at each location is calculated using the formula T = (ρ×D) / (1 +ρ) to generate a continuous tension field that quantifies the intensity of ethical conflict, thus forming an ethical tension map.

[0078] Specifically, for each viewpoint or disagreement, an ethical feature vector is extracted based on the weight distribution of the ethical principles involved. Principal Component Analysis (PCA) is used to identify the two most significant dimensions of variation in the data and uses them as coordinate axes in the ethical space. For example, the first principal component typically corresponds to the individual autonomy-group interest axis (x-axis), and the second principal component corresponds to the rule compliance-consequence consideration axis (y-axis). By projecting each feature vector onto the principal components, its coordinates in the two-dimensional ethical space are obtained. After obtaining the coordinates, the ethical space is divided into a grid, and for each grid point (i, j), the local viewpoint density ρ(i, j) and the local disagreement intensity D(i, j) are calculated. ρ(i, j) can be obtained by summing Σ over all viewpoints. k exp(-d 2 k / 2σ 2 ) to calculate, where d k Let D(i,j) be the distance from the k-th viewpoint to the grid point. D(i,j) can be calculated by summing Σ over all divergences. m conflict m ×exp(-d 2 m / 2σ 2 ) to calculate, where conflict mLet represent the intensity or crystallinity of the m-th divergence. Local tension is calculated using the formula T(i,j) = ρ(i,j) × D(i,j) / (1 + ρ(i,j)). This formula represents a local divergence intensity modulated by viewpoint density; that is, high tension only occurs when divergences appear in areas of dense viewpoints. This avoids the undue emphasis on isolated divergences, allowing the generated tension map to more accurately reflect the core areas of ethical conflict.

[0079] This embodiment elevates the final output of the game theory discussion from a simple list of viewpoints to a new level with in-depth diagnostic value and intuitive decision support capabilities, solving the problems of limited output and insufficient insight in existing methods. By assessing the stability of the time series of the evolutionary argument intensity matrix and combining it with the calculation of multi-dimensional crystallinity indicators (including time stability, intensity symmetry, etc.), it can not only identify disagreements but also crystallize them, that is, quantify their stability and root depth. Principal component analysis (PCA) maps all viewpoints and disagreements into a data-driven ethical space, and calculates a continuous tension field by combining viewpoint density and disagreement intensity. In specific medical ethics analysis scenarios, its value to decision-makers is enormous: the crystallized disagreement set can clearly tell the ethics committee which opposing viewpoints are irreconcilable structural contradictions that require risk management; while the ethical tension map can clearly see where the core issues of the entire ethical dilemma lie, thus allowing for focused discussion and intervention on high-tension areas, improving the pertinence and efficiency of decision-making.

[0080] In a preferred embodiment, the excited-state decay mechanism also includes a regeneration mechanism corresponding to the decay. This mechanism simulates the phenomenon in real debates where old arguments regain attention due to new circumstances. Specifically, the system continuously monitors whether there are significant changes in the game's context, for example, by calculating the KL divergence of the argument topic distribution within a continuous time window. When the KL divergence exceeds a preset threshold (e.g., 0.5), a regeneration event is triggered, allowing older, weaker arguments to regain some energy. The energy update formula can be E. regen = E current +0.3×E0×novelty score E current For current energy, novelty score It measures how well the old argument fits the new context.

[0081] Furthermore, to ensure the macroscopic stability of the game and prevent abnormal increases or excessively rapid decays in the total system energy due to coupling flow, a system energy conservation constraint mechanism is introduced. Specifically, in the coupling flow calculation at each time step, an energy conservation constraint is applied, such as Σ. i Ei (t) = Σ i E i (0)×exp(-γ avg ×t), ensuring that the total energy of the system follows the preset average decay rate γ. avg The constraint can be solved and applied using the Lagrange multiplier method, thus ensuring the macroscopic stability of the game.

[0082] Furthermore, a unique argument credibility radar chart can be generated during the report generation phase. This chart is a unique and valuable output directly derived from resonance suppression. Specifically, based on resonance suppression records generated during the game, it is used to demonstrate the stability performance of each argument in the game, i.e., the degree to which it is suppressed. For example, an argument with a very low stability score in the radar chart means that it frequently participates in circular references and triggers resonance in the game, thus being suppressed by the system, and its independent credibility may need to be carefully assessed. This chart can intuitively reflect which arguments behave improperly in the game, enhancing the transparency and credibility of the final analysis report.

[0083] In a preferred embodiment, to improve the accuracy of consensus point identification, statistical significance is determined for the identified resonant frequencies during cross-spectral analysis. Specifically, after calculating the phase-locked value (PLV), a permutation test is used to calculate the p-value for each synchronization frequency. For example, the null hypothesis distribution of the PLV can be established by randomly shuffling the phase 1000 times. Only when the calculated p-value is less than the significance level (e.g., p < 0.01) is the resonant frequency considered a statistically significant, non-random synchronization event, and its corresponding argument is confirmed as a potential fusion point. This embodiment enhances the scientific rigor and reliability of consensus point identification, representing a specialized step distinct from simple signal processing.

[0084] In a preferred embodiment, to deepen the understanding of structural disagreements, argument mining techniques are employed in the root cause tracing step to extract the internal structure (premise-reasoning-conclusion) of stable opposing arguments. Based on this, by comparing their structural differences, the root cause of the disagreement is automatically categorized into one of three types: value premise disagreement, factual perception disagreement (e.g., differing assessments of the probability of cure by both parties), or reasoning path disagreement (e.g., both parties start from the same premises but arrive at conflicting conclusions through different logical deductions). This provides deeper insights for ethical analysis, enabling the output of classification results with clear diagnostic significance, rather than simply identifying points of disagreement.

[0085] In addition to the credibility radar chart, another preferred implementation during the report generation phase may include generating a divergence evolution path diagram. Based on the results of automatic tracing of the root causes of the divergence, this diagram visually illustrates, in a tree-like or graphical manner, how the core root divergence (such as a divergence of value premises) evolves and expands into specific, superficial positional oppositions through a series of intermediate arguments. This visualization helps users clearly understand the source of the conflict, providing a more in-depth basis for decision-making in finding reconciliation solutions or managing risks.

[0086] According to another aspect of this application, a medical ethics analysis report generation system based on multi-model game theory is provided, the system comprising at least:

[0087] Data Input and Preprocessing Unit: Configured to receive and structure medical ethics-related information to be analyzed from one or more sources. This information may include, but is not limited to, medical case descriptions, relevant medical background knowledge, current legal and regulatory texts, ethical guidelines and standards, socio-cultural background information, and user-defined analytical focuses. Preprocessing includes data cleaning, key information extraction (such as ethical elements, stakeholders, and core conflict points), knowledge representation (such as constructing a temporary case knowledge graph or associating it with an existing knowledge base), and preliminary question classification.

[0088] Multi-model Independent Analysis and Role-Based Units: These units comprise at least two (and often more) analytical agents driven by large language models (LLMs). Each agent can be configured as isomorphic (using the same underlying LLM but with different initializations or fine-tunings) or heterogeneous (using different underlying LLMs). The core feature is the ability to assign specific ethical analysis roles or perspectives to each agent or group of agents (e.g., patient rights representative, legal compliance officer, social impact assessor, etc.). These role-based agents receive pre-processed information and specific analytical instructions, independently perform preliminary interpretations of the input information, and generate preliminary analytical opinion fragments containing core arguments, preliminary evidence, and ethical judgments based on their assigned roles.

[0089] Inter-model Game Decision-Making Unit: This unit is responsible for designing and implementing a structured interactive game mechanism between analytical agents. This mechanism includes at least: a communication and coordination platform for managing orderly speech and information sharing among agents; a question generation and transmission mechanism: allowing one or more agents (or a dedicated coordinating / critical agent) to initiate structured questions or challenges to the target agent based on differences in the initial analytical opinions of other agents, logical consistency checks, sufficiency of evidence assessments, or ethical compliance considerations; a defense and argument reinforcement mechanism: requiring the challenged agent to defend, clarify, supplement evidence, revise arguments, or adjust its position under reasonable questioning; this process may involve multiple rounds of interaction; and a consensus formation (or disagreement clarification) mechanism: through tracking viewpoint evolution, evaluating argument strength, and detecting semantic consistency, after the game process reaches preset conditions (such as convergence, reaching the maximum number of rounds, or key disagreements being fully exposed), determining the set of consensus viewpoints reached among agents on core ethical issues, or clearly identifying and recording irreconcilable major disagreements and their respective argumentation bases.

[0090] The Ethics Integration and Report Generation Unit: Based on the comprehensive analysis results output by the inter-model game decision-making unit (consensus viewpoints, reinforced arguments, clear disagreements and their reasons), combined with the system's built-in core medical ethics principles knowledge base and multi-dimensional analysis framework (such as clinical, research, new technology, public health, and cultural and social dimensions), and referring to a structured report template, this unit automatically organizes the content and generates a logically clear, complete, and well-reasoned draft of a medical ethics analysis report. This unit is also responsible for the automatic organization and formatting of references.

[0091] Solution Formulation and Evaluation Unit: Based on the key ethical issues and challenges identified in the ethical analysis report, this unit is responsible for inspiring one or more LLMs (which can reuse analytical agents or be configured with specialized solution design agents) to generate targeted and actionable solution recommendations. These recommendations should include elements such as objectives, specific measures, implementation paths, expected effects, and potential risks. The generated solutions themselves can also be fed back to the inter-model game decision-making unit for a new round of ethical evaluation and optimization.

[0092] Furthermore, a method for generating medical ethics analysis reports based on multi-model game theory is proposed. This method corresponds to the aforementioned system and includes at least the following steps:

[0093] Receive and preprocess medical ethics-related information submitted by users or actively acquired by the system;

[0094] The preprocessed information and analysis instructions are distributed to multiple role-based LLM-driven analysis agents, enabling them to perform independent analysis and output preliminary analysis opinions.

[0095] Initiate and manage the game-theoretic decision-making process between intelligent agents, which includes at least one round of interaction consisting of question generation, response, and argument reinforcement.

[0096] After the game process is over, extract and integrate the findings to form a consensus-based analytical conclusion on the core ethical issues or a comprehensive opinion that includes clear disagreements.

[0097] Based on the preset core medical ethics principles, multi-dimensional analysis framework, and analysis conclusions after game theory integration, a structured medical ethics analysis report is automatically generated.

[0098] Based on the generated ethical analysis report, further inspiration can be generated and evaluated for proposed solutions to ethical dilemmas.

[0099] In order to concretize the core calculation process of the present invention and make it reproducible to those skilled in the art, this embodiment provides a specific numerical calculation case on the dynamic evolution of the argument intensity, which will run through the entire process of initial energy level allocation, excited state decay and coupling network flow.

[0100] Suppose a medical ethics case containing three core initial arguments: Argument A: The patient's wish to forgo treatment should be respected, and the decision should be made by the patient's role (weight w). role Argument B (=1.2) is a principle-based argument. Argument B: The feelings of the family should be considered, and active treatment should continue, with the family's role (weight w) taking precedence. role Argument C (weight w) is proposed by a doctor (weight w = 1.0) and belongs to the category of emotional argumentation. role Argument A (=1.2) is an obligatory argument. A three-dimensional completeness score is applied to the three arguments, and the initial energy level E0 is calculated. Assume the following scoring results: Argument A: Clear premise (P=0.9), clear logic (L=0.9), relevant conclusion (R=0.9). Argument B: Clear premise (P=0.7), weak logic (L=0.6), relevant conclusion (R=0.8). Argument C: Clear premise (P=0.8), clear logic (L=0.9), relevant conclusion (R=0.8). Calculate the overall quality score Q. total = (P×L×R) (1 / 3) Q total_A = (0.9 * 0.9 * 0.9) (1 / 3) = 0.90; Q total_B = (0.7 * 0.6 * 0.8) (1 / 3) ≈ 0.70; Q total_C = (0.8 * 0.9 * 0.8) (1 / 3) ≈ 0.83. Calculate the score Q after adjusting for role weights. adjusted = Qtotal ×w role Q adjusted_A = 0.90 * 1.2 = 1.08; Q adjusted_B = 0.70 * 1.0 = 0.70; Q adjusted_C = 0.83 * 1.2 ≈ 1.00. The initial energy level E is assigned using the Softmax function for normalization. 0_i = E total × exp(Q adjusted_i ) / Σ j exp(Q adjusted_j Let the total energy of the system be E. total =10.0: exp(1.08) ≈ 2.94, exp(0.70) ≈ 2.01, exp(1.00) ≈ 2.72; denominator Σ = 2.94 + 2.01 + 2.72 = 7.67; E 0_A = 10.0 * 2.94 / 7.67 ≈ 3.83; E 0_B = 10.0 * 2.01 / 7.67 ≈ 2.62; E 0_C = 10.0 * 2.72 / 7.67 ≈ 3.55 (Verification: 3.83 + 2.62 + 3.55 = 10.0, which satisfies the total energy constraint).

[0101] Assume semantic distance and coupling strength K between arguments ij (For simplicity, let's assume symmetry) as follows: A and C (respectful will vs. doctor's obligation) are semantically close and highly conflicting: semantic distance (A, C) = 0.4, K AC =0.25. The semantic distance between A and B (patient's wishes vs. family's emotions) is moderate: semantic distance (A, B) = 0.7, K AB =0.15. The semantic distance between B and C (family emotions vs. doctor's obligations) is significant: semantic distance (B, C) = 1.0, K BC =0.08.

[0102] Suppose that 10 seconds after the game begins, argument A is cited once by another model (n). A =1). At the current moment (t=10s), the strength of each argument has slightly decreased due to its own decay, let's assume: E A (10) = 3.50, E B (10) = 2.50, E C(10) = 3.30. Excitation and decay rate update (for argument A): Calculate the excitation gain coefficient α. A = 0.15×exp(-0.1×1) ≈ 0.136. Prove that the instantaneous energy level of A after being excited is E. excited =E A (10)×(1 +α A = 3.50 * (1 + 0.136) ≈ 3.98. Update the argument for the nonlinear decay rate γ of A. A Assuming its basic attenuation rate γ 0_A =0.1, β=0.05, and the current citation frequency and interval factor are 1.1 and 1.05, respectively. Then the new γ A = 0.1×(1 + 0.05×1 2 )×1.1×1.05 ≈ 0.121. Coupled flow calculation (within the next time step dt=0.1s): Calculate the coupling effect on A: ΔE A_coup = [K AB (E B -E A ) + K AC (E C -E A )] × dt = [0.15(2.5-3.98) + 0.25(3.3-3.98)] × 0.1 ≈ -0.040. Calculate the coupling effect on B: ΔE B_coup = [K AB (E A -E B ) + K BC (E C -E B )]×dt = [0.15(3.98-2.5) + 0.08(3.3-2.5)]×0.1≈+0.029. Calculate the coupling effect on C: ΔE C_coup =[K AC (E A -E C )+K BC (E B -E C )]×dt =[0.25(3.98-3.3) + 0.08(2.5-3.3)]×0.1≈+0.011. Total intensity change calculation: E A (10.1)=E excited ×exp(-γ A ×dt)+ΔE A_coup= 3.98×exp(-0.121×0.1) - 0.040≈3.94 - 0.040 = 3.90. E B (10.1) = E B (10)×exp(-γ B ×dt) +ΔE B_coup ≈2.50×exp(-γ B ×0.1) + 0.029 (assuming γ) B (Unchanged). E C (10.1) = E C (10)×exp(-γ C ×dt) +ΔE C_coup ≈3.30×exp(-γ C ×0.1) + 0.011 (assuming γ) C (Unchanged).

[0103] This case clearly demonstrates the entire process by which argument A's strength instantly increases after being cited, and then evolves under its own faster decay and the coupling effect from other arguments, thus verifying the operability and specific effects of the present invention.

[0104] This invention addresses the problem of modeling argument fatigue by employing an excited-state decay mechanism, particularly by introducing a dynamic evolution formula with the number of citations and the time since the last citation as core variables. Mathematically, it ensures that the influence of any argument wears down over time and with repeated use, consistent with real-world debate intuition. This effectively prevents a single viewpoint from dominating the discourse in medical ethics discussions through simple repetition, guaranteeing dynamic fairness in the game process. An adaptive damping mechanism resolves the issue of uncontrolled resonance. Through a closed-loop monitoring-computation-inhibition mechanism, it can identify vicious circular citation structures in the network in real time and apply smooth growth damping only to arguments involved. Working in conjunction with the excited-state decay mechanism—the former managing the lifecycle of individual arguments and the latter managing inappropriate interactions between arguments—the two jointly maintain the macroscopic stability of the entire game system and the reliability of the analysis results. Furthermore, by employing divergence crystallization and ethical tension map generation methods, it enhances the depth and usability of the analysis results while addressing procedural issues. Building upon a more realistic game outcome that has undergone dynamic decay and resonance suppression, further diagnostic information and decision support tools are provided to users (such as hospital ethics committees) by quantifying the stability of disagreements (i.e., crystallinity) and visualizing the core areas of ethical conflict (i.e., tension maps). This constitutes a systemic transcendence of the entire chain of defects, from process distortion to superficial results.

[0105] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A method for generating medical ethics analysis reports based on multi-model game theory, characterized in that, include: Analyze original medical ethics case texts to construct a structured set of ethical elements and a set of role-based model configurations; A multi-model game was played on the set of structured ethical elements and the set of role-based model configurations. The strength of the argument was dynamically evolved through the excitation-state decay mechanism, and the evolutionary argument strength matrix, game trajectory data and resonance suppression records were obtained. By analyzing the evolutionary argument intensity matrix and game trajectory data, we can explore the structural divergences of viewpoint fusion and crystallization, and obtain the fused viewpoint set, the crystallized divergence set, and the ethical tension map. By integrating and fusing viewpoint sets, crystallized divergence sets, ethical tension maps, and resonance inhibition records, a medical ethics analysis report is generated. The evolutionary argument strength matrix, game trajectory data, and resonance suppression records were obtained, including: Each role-based model generates an initial argument based on a structured set of ethical elements, and assigns an initial energy level to each argument through a completeness score, thereby constructing an initial argument viewpoint library and an argument association matrix that represents the semantic relationship between viewpoints; In multi-round game interaction, the excited state decay model, coupled oscillator network and adaptive damping mechanism are integrated and applied to dynamically evolve the argument strength in the initial argument viewpoint base; Among them, the excited-state decay model simulates the fatigue effect of the argument based on the citation history and time of the argument, the coupled oscillator network realizes the intensity redistribution among the related viewpoints based on the argument correlation matrix, and the adaptive damping mechanism suppresses the resonance amplification caused by circular references. The final result of the dynamic evolution is recorded and output as an evolutionary argument strength matrix, game trajectory data, and resonance suppression record.

2. The method according to claim 1, characterized in that, The excited-state decay model simulates and demonstrates the fatigue effect in the following way: For any argument in the initial argument base, its argument strength E(n,t) = E0 × (1 + α×n) × exp(-γ(n)×t) + E base Where E0 is the initial energy level of the argument, n is the cumulative number of times the argument has been cited in the game process, t is the evolution time since the last citation, γ(n) is the nonlinear decay rate that adaptively adjusts with the number of citations n, and α is the preset excitation gain coefficient. base This is the preset ground state energy level.

3. The method according to claim 2, characterized in that, The nonlinear decay rate γ(n) is adaptively adjusted in the following manner: The basic attenuation rate γ0 is determined based on the argument type preset in the initial argument viewpoint library. Calculate the citation frequency f from the citation history pattern data of the argument. cite Standard deviation σ of the reference interval distribution interval ; The formula γ(n) = γ0 × (1 + β×n) 2 ) × (1 + 0.5×log(1+f cite )) × exp(σ interval / σ max The nonlinear attenuation rate γ(n) is calculated; where β is a preset coefficient controlling the influence of the number of citations on the intensity, and σ max This is the standard deviation of the maximum citation interval used for normalization.

4. The method according to claim 1, characterized in that, Coupled oscillator networks achieve the redistribution of intensity among associated viewpoints in the following way: For any argument i in the initial argument base, the intensity change ΔE caused by coupling effect i The interaction effects of all other arguments j are summed and calculated. DE i = S j K ij × (E j - E i ) × Φ(semantic distance (i,j)); Among them, E i With E j K represents the current argument strength of arguments i and j, respectively. ij To demonstrate the coupling strength derived from the correlation matrix, semantic distance (i, j) represents the semantic distance between i and j, and Φ is a preset function that maps this semantic distance to coupling weights.

5. The method according to claim 4, characterized in that, Semantic distance distance The calculation methods for (i, j) include: On a pre-constructed medical ethics concept graph, the shortest path on the graph between the core concepts of argument i and j is calculated to obtain the concept distance component; By analyzing the text content of arguments i and j, we extract the stance vector and the core object of the argument, and calculate the stance similarity component and the overlap component of the argument object. The semantic distance is comprehensively assessed by weighting and summing the conceptual distance component, the positional similarity component, and the overlap component of the argumentative objects. distance (i, j).

6. The method according to claim 1, characterized in that, The adaptive damping mechanism suppresses resonant amplification in the following ways: Real-time monitoring of strongly coupled loops within the argumentation network, and calculation of an index reflecting the current loop strength for each argumentation participating in the loop; When the cyclic strength index of any argument exceeds a preset dynamic threshold, the argument is activated and adaptive damping is applied. Adaptive damping is integrated into the strength evolution calculation of this argument to suppress its aberrant strength amplification; The triggering and effects of the suppression process are recorded in the resonance suppression log.

7. The method according to claim 6, characterized in that, The calculation method for adaptive damping is as follows: Through formula D i (t) = D0 × [1 + tanh(μ×(C i (t) - C threshold The adaptive damping D was calculated. i (t); Where D0 is the preset base damping coefficient, and μ is the sensitivity parameter controlling the smoothness of damping activation. This formula utilizes the properties of the hyperbolic tangent function tanh to ensure that when the cyclic strength index C... i (t) approaches and exceeds the dynamic threshold C threshold When, adaptive damping D i The value of (t) increases smoothly.

8. The method according to claim 1, characterized in that, Obtain a fusion of perspectives, including: Cross-spectral analysis was performed on the time series of interactive argument intensity in game trajectory data. By calculating the phase lock value, arguments that undergo phase synchronization at a specific resonance frequency were identified. By tracing and merging the arguments that are in phase synchronization, and extracting common deep-seated value claims through semantic clustering, a set of integrated viewpoints is constructed.

9. The method according to claim 1, characterized in that, Obtaining crystallized branch sets includes: By evaluating the stability of the evolutionary argument strength matrix, we can identify argument pairs that maintain a stable adversarial state in the later stages of the game. For each identified stable adversarial state, the root cause of the divergence is traced, and a multidimensional crystallinity index that quantifies the stability of the divergence is calculated. A crystallized divergence set containing the argument pair and the corresponding multidimensional crystallinity index is constructed.

Images