A constraint-based decision path backtracking method based on time sequence argument graph

Abstract

The application discloses a constraint decision path reverse tracing method based on a time sequence argumentation graph, and belongs to the technical field of the cross of artificial intelligence and knowledge graph; first, the expert speech content and the final decision conclusion of the review meeting are collected, argumentation elements and mutual relations are extracted, and a time sequence argumentation graph containing four types of nodes and five types of edges is constructed; second, the node influence potential, the time sequence propagation influence and the cumulative influence intensity of each upstream node are calculated through a time sequence influence propagation model; then, the final conclusion node is taken as a search starting point, a reverse breadth-first search is performed by applying three constraints, and a tracing path subgraph is screened; subsequently, the life cycle management of objections, expert responsibility attribution and four-chain collaborative explanation and analysis are cooperatively carried out; finally, a structured decision tracing and responsibility attribution report is output; the application realizes time sequence argumentation modeling and accurate reverse tracing, can quantize expert contribution and objection influence, has strong decision explainability, and is suitable for scenes such as scientific research review and bidding and bid evaluation.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of artificial intelligence and knowledge graph, specifically involving a method for reverse tracing of constraint-based decision paths based on temporal argument graphs. Background Technology

[0002] In complex review meetings such as scientific research project evaluation and bidding evaluation, the review process involves multiple experts sequentially expressing their views, providing evidence, and raising objections over time, ultimately converging to form a decision. The traceability and explainability of the decision-making process are core requirements for ensuring the fairness of the review and clarifying review responsibilities.

[0003] Existing technologies have several shortcomings: traditional review management systems only record the final conclusions and voting results, lacking reverse traceability capabilities and unable to explain the formation process of the conclusions; blockchain-based review evidence storage schemes can only guarantee the immutability of process data, providing only evidence storage functions without offering causal reasoning and impact analysis capabilities; knowledge graph-based argumentation modeling schemes only store argumentation relationships in a structured manner, without incorporating a temporal dimension, lacking constrained reverse search capabilities, and unable to accurately locate the impact paths that play a key role in decision-making; graph neural network reasoning schemes are black-box models, with uninterpretable output results and unable to provide clear traceability paths; rule-based expert systems rely on predefined rule reasoning, lacking flexibility and unable to adapt to complex and diverse review scenarios.

[0004] In summary, existing technologies cannot simultaneously achieve time-series argumentation impact modeling, constraint-based precise reverse tracing, full lifecycle management of objections, quantitative attribution of expert contributions, and multi-dimensional collaborative explanation, making it difficult to meet the decision tracing needs in complex review scenarios. Summary of the Invention

[0005] To address the problems mentioned in the background art, this invention provides a method for reverse tracing of constrained decision-making paths based on temporal argumentation graphs, thereby solving the technical problems in the prior art such as the lack of temporal dimension in argumentation modeling, inability to accurately reverse trace decision-making paths, inability to track opposing opinions throughout their entire lifecycle, inability to quantify expert contributions, and insufficient interpretability of decisions.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A constraint-based decision path reverse tracing method based on temporal argumentation graphs includes the following steps:

[0008] S1: Collect the speeches and final decision conclusions of multiple experts in the review meeting, extract the viewpoints, evidence, conclusions, objections and the relationship between them through natural language processing, and construct the time sequence argumentation graph G=(V,E,T,W,Φ,Δ);

[0009] The node set V includes four types: opinion nodes, evidence nodes, final conclusion nodes, and opposing nodes; the directed edge set E includes five types: supporting edges, opposing edges, rebuttal edges, citation edges, and derived edges; T is the node timestamp function, used to characterize the time of generation of each node; W is the edge weight function, used to characterize the strength of each argument relationship; Φ is the expert authority function, used to characterize the degree of authority of each expert; Δ is the time decay function.

[0010] S2: Based on the temporal argumentation graph, perform temporal influence propagation model calculation, sequentially calculate the influence potential of each node, the temporal propagation influence of each directed edge, and the cumulative influence intensity of each node on the final conclusion node, forming an influence intensity calculation result that includes influence potential, temporal propagation influence, and cumulative influence intensity;

[0011] S3: Starting from the final conclusion node, perform a triple-constraint reverse breadth-first search based on the influence intensity calculation results. During the search process, simultaneously apply strong influence constraints, causal integrity constraints, and time consistency constraints to filter out source path subgraphs that satisfy all triple constraints and the corresponding influence intensity cache.

[0012] S4: Based on the source tracing path subgraph and impact intensity cache, the life cycle management analysis of objections, the expert responsibility attribution analysis, and the generation of four-chain collaborative interpretation are performed in a coordinated manner to obtain the four-chain collaborative interpretation results, which include the key viewpoint chain, the evidence support chain, the objection handling chain, and the expert contribution chain.

[0013] S5: Integrate the results of the four-chain collaborative explanation with the corresponding quantitative indicators, and output a structured decision tracing and responsibility attribution report.

[0014] The beneficial effects of this application are:

[0015] 1. This application uses a time decay function to model and demonstrate the temporal propagation process of the influence, which satisfies the monotonic decay of the influence and the order preservation of time decay. It conforms to the cognitive law of "first impressions and decreasing" in the review scenario, has a clear physical meaning, and fills the gap in the existing technology that has not systematically integrated the temporal dimension.

[0016] 2. The triple constraint reverse search of this application can significantly reduce the invalid search space, and the triple constraint ensures the causal integrity and logical legality of the tracing path.

[0017] 3. This application precisely calculates expert contributions from four dimensions: viewpoint, evidence, objections, and timeline, thereby achieving a reasonable allocation of responsibility and elevating expert contributions from unquantifiable to precise four-dimensional quantification. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the process of this application;

[0019] Figure 2This is a schematic diagram of the time decay function. Detailed Implementation

[0020] To facilitate understanding of the technical content of this invention by those skilled in the art, the invention will be further described in detail below with reference to the accompanying drawings and specific examples. It should be understood that the specific examples described herein are merely illustrative and not intended to limit the scope of the invention.

[0021] A method for reverse tracing of constrained decision paths based on temporal argumentation graphs, such as... Figure 1 As shown, the specific steps are as follows:

[0022] S1: Construction of the temporal argumentation diagram;

[0023] S1.1: Data collection; collecting the statements of all experts at the review meeting and the final decision conclusions reached at the meeting.

[0024] S1.2: Argument Extraction; Four types of argument elements are extracted from the speech content using natural language processing technology, corresponding to the construction of four types of nodes:

[0025] Opinion nodes: Represent the review opinions, judgments, and evaluations put forward by experts. All opinion nodes constitute a subset V of opinion nodes. opinion ;

[0026] Evidence nodes: These represent the supporting materials for a viewpoint. All evidence nodes constitute a subset V of evidence nodes. evidence ;

[0027] Opposing nodes: Represent the opposing opinions and questions raised by experts; all opposing nodes constitute a subset V of opposing nodes. opposition ;

[0028] Final conclusion node: corresponds to the final decision conclusion of the review, and belongs to the subset V of conclusion nodes. conclusion , which is the final result node of the entire argument diagram.

[0029] Each node is associated with a corresponding timestamp, recording the time when the argument element was generated; at the same time, a node-expert mapping function expert:V→ExpertSet is defined, representing the expert who proposed each node.

[0030] S1.3: Extraction of Argument Relationships; Extract five types of argument relationships between nodes and construct five types of directed edges accordingly:

[0031] Support edges: Represent the positive supporting relationships between viewpoints and between evidence and viewpoints. All support edges constitute a subset E. support ;

[0032] Opposing edges: Represent the negation relationship between opposing opinions and viewpoints; all opposing edges constitute a subset E. oppose ;

[0033] Refutation edges: Represent the rebuttal response of a viewpoint to opposing opinions; all refutation edges constitute a subset E. rebut ;

[0034] Reference edges: Represent the relationship between an opinion and evidence; all reference edges constitute a subset E. cite ;

[0035] Derived edges: Represent the logical derivation relationship from the upstream premise node to the final conclusion node. All derived edges constitute a subset E. derive .

[0036] S1.4: Formal definition of the temporal argument graph; the final construction of the temporal argument graph is a six-tuple G=(V,E,T,W,Φ,Δ), with each element fully defined as follows:

[0037] Node set: It includes four categories: opinion nodes, evidence nodes, final conclusion nodes, and opposing nodes;

[0038] Set of directed edges: It includes five categories: supporting edges, opposing edges, rebuttal edges, referencing edges, and derived edges;

[0039] Node timestamp function: T(v) represents the time when node v is generated;

[0040] Edge weight function: , Characterizes the strength of each argument relationship;

[0041] Expert authority function: This is used to characterize the level of authority of each expert;

[0042] Time decay function: ,like Figure 2 As shown, this is used to characterize the degree to which the influence of the argument decays over time.

[0043] The edge weight function is determined based on the edge type and temporal context: the weight of a supporting edge is proportional to the confidence level of the viewpoint node it connects to and the authority of the expert who proposed the viewpoint node; the weight of an opposing edge is proportional to the strength of the argument of the opposing node it connects to; the weight of a rebuttal edge is proportional to the specificity of the rebuttal relationship it carries; the weight of a citation edge is proportional to the reliability of the evidence node it connects to; and the weight of a derived edge is proportional to the logical consistency between the final conclusion node it connects to and the upstream premise node.

[0044] S2: Calculation of the Time-Series Effects Propagation Model (TIPM);

[0045] Based on the constructed temporal argumentation graph, a temporal influence propagation calculation is performed to quantify the degree of influence of each node on the final conclusion, specifically including:

[0046] S2.1: Calculate the node influence potential; the node influence potential is the potential influence of the node itself, determined by expert authority, content confidence, and node type. For any node v∈V, its influence potential I(v) is calculated using the following formula:

[0047] ;

[0048] in, To determine the expert authority of this node, The confidence level is a built-in label for the node. The node type weights are set as follows: opinion nodes are assigned 1.0, evidence nodes are assigned 0.8, opposing nodes are assigned 0.6, and conclusion nodes are assigned 1.2.

[0049] S2.2: Calculate the temporal propagation effect; For any directed edge e=(u,v)∈E, the formula for calculating the temporal propagation effect P(u→v) from the source node u to the target node v is:

[0050] ;

[0051] Where W(e) is the weight of the directed edge; Δ(T(u),T(v)) is the time decay function, which adopts an exponential decay form:

[0052] ;

[0053] In the formula, λ>0 is the decay rate hyperparameter, t u Let t be the timestamp of the source node u. v The timestamp of the target node v.

[0054] S2.3: Calculate the cumulative influence intensity; For an upstream node v that has a directed influence path to the final conclusion node, its cumulative influence intensity I(v,c) on the final conclusion node c is a weighted aggregation of path influence and direct influence, calculated using the following formula:

[0055] ;

[0056] Where c∈V conclusion , The path influence strength is the sum of the products of the temporal propagation influences on all paths from node v to the final conclusion node c:

[0057] Π(v,c) is the set of all directed paths from v to c;

[0058] To directly affect the strength, that is, the maximum value of the temporal propagation influence of all direct predecessor nodes from node v to the final conclusion node c: ;

[0059] This is the path-direct equilibrium coefficient, used to adjust the weight ratio of path aggregation impact to direct impact.

[0060] S3: Triple Constraint Search Mechanism (TCSM) reverse tracing;

[0061] Starting from the final conclusion node, a reverse breadth-first search with triple constraints is performed to select the source path subgraph that plays a key role in decision formation. The specific process is as follows:

[0062] S3.1: Initialization: Add the final conclusion node to the search queue and the visited set, and initialize the source path node set, edge set, and influence strength cache.

[0063] S3.2: Triple Constraint Filtering: Traverse all incoming edges of the current node in reverse order, and check each edge against the triple constraints. Only nodes and edges that satisfy all constraints are retained.

[0064] (1) Strong Influence Constraint: Only nodes whose cumulative influence on the final conclusion node is not less than the preset node threshold θinf, and directed edges whose temporal propagation influence is not less than the preset edge threshold θedge are retained. Weakly influential nodes and edges are filtered out, and the core influence path is focused. Specifically, this is expressed as:

[0065] ;

[0066] ;

[0067] in and These are the set of nodes and the set of edges in the tracing path, respectively.

[0068] (2) Causal integrity constraint: The sequence of adjacent triplet node types on the reverse search path must belong to the predefined legal causal pattern library and not to the illegal causal pattern library, ensuring the legality of the path's causal logic. Specifically, this is expressed as:

[0069] ;

[0070] The legitimate causal triplet pattern library M is defined as follows:

[0071] ;

[0072] At the same time, the following illegal causal models are explicitly prohibited:

[0073] ;

[0074] (3) Time consistency constraint: The generation time of the source node corresponding to each edge on the reverse search path is no later than the generation time of the target node, ensuring temporal causality; and the time interval between adjacent nodes does not exceed the preset maximum allowable interval. To filter weakly associated paths with excessively long time spans, the following expression is used:

[0075] ;

[0076] .

[0077] S3.3: Search Optimization Measures; To improve the search efficiency of large-scale graphs, three optimization measures are adopted:

[0078] (1) Impact strength memoization search: The cumulative impact strength is calculated incrementally using dynamic programming, and the impact strength results of already calculated nodes are cached to avoid duplicate calculations. Specifically:

[0079] .

[0080] in This is a value that has already been cached in the previous path calculation. For nodes The set of successor nodes. This optimization avoids redundant path enumeration and reduces the time complexity of path aggregation computation from... Down to .

[0081] (2) Causal Pattern Library Pruning: During the search process, the current path prefix is ​​checked. If an illegal causal pattern is matched, the search of that branch is terminated, reducing the search space. Specifically:

[0082] use Pruning the search space: If any illegal pattern is detected in the current path prefix, the search for that branch is terminated immediately.

[0083] Pruning decision function: .

[0084] (3) Hierarchical Time Index: A timestamp-based bucket index structure is pre-built. During the search, only the buckets whose time is no later than the current node are retrieved, reducing the complexity of inbound edge retrieval. Specifically:

[0085] A pre-built timestamp-based Bucket index structure is used to discretize nodes by time. One bucket:

[0086] ;

[0087] During BFS input edge traversal, only time buckets are searched. The nodes in the middle will reduce the complexity of incoming edge retrieval from Down to .

[0088] S3.4: Output results; After traversal, output the source path subgraph that satisfies all triple constraints, as well as the corresponding influence intensity cache.

[0089] S4: Multi-dimensional collaborative analysis;

[0090] Based on the source path subgraph and influence intensity caching, the analysis of three modules is performed collaboratively:

[0091] S4.1: Objection Lifecycle Management (OLM) Analysis; Tracking and quantifying all objection nodes throughout their entire lifecycle in the tracing path:

[0092] (1) Five-state lifecycle modeling: The lifecycle of objections is modeled as a finite state machine containing five states: proposed, under discussion, refuted, adopted, and pending processing, expressed as:

[0093] The meanings of each state are shown in Table 1:

[0094] Table 1

[0095]

[0096] The state transition diagram can be formally represented as:

[0097] ;

[0098] in For the event alphabet, , Let be the set of final states.

[0099] (2) Calculate the objection handling completion rate (OHCR):

[0100] ;

[0101] OHCR is used to quantify the extent to which objections are adequately addressed during the review process. A higher OHCR indicates a higher proportion of objections have been adequately responded to.

[0102] (3) Calculate the Opposing Opinion Adoption Impact Index (OAI): This quantifies the proportion of influence of adopted opposing opinions in the decision-making process. The formula is as follows: ;

[0103] in For opposing nodes Regarding the conclusion The strength of influence, where the denominator is the total strength of influence of all viewpoints on the conclusion. To prevent the use of tiny constants that divide by zero, OAI quantifies the proportion of influence of dissenting opinions in actual decision-making; a larger value indicates a more significant corrective effect of dissenting opinions on the decision.

[0104] S4.2: Expert Responsibility Attribution (EAA) Analysis; quantifying each expert's contribution to the final conclusion from four dimensions to achieve precise responsibility allocation:

[0105] (1) Contribution of viewpoint influence: The sum of the cumulative influence of all viewpoints raised by experts on the final conclusion, as shown in the formula: ;

[0106] (2) Evidence Support Contribution: The weighted sum of the influence of all evidence points provided by experts on the final conclusion and the reliability of the evidence, as shown in the formula:

[0107] ;reliability(v) represents the reliability of evidence node v.

[0108] (3) Contribution of dissenting opinions: The weighted sum of the strength of the impact of the dissenting opinions raised by experts on the final conclusion and the corresponding adoption polarity, as shown in the formula: ;

[0109] Where sign(o) is the polarity function of the objection, which takes a value of +1 when it is adopted, a value of -1 when it is refuted, and a value of 0 when it is unresolved.

[0110] (4) Time-series contribution factor: Considering the time-sensitivity weight of expert speaking time, the formula is: ;

[0111] The experts' overall responsibility attribution score was obtained by combining contributions across the four dimensions:

[0112] ;

[0113] in For dimension weights, satisfying The default weighting scheme is The overall scores were normalized to obtain the relative responsibility coefficients of each expert:

[0114] .

[0115] S4.3: Four-chain collaborative interpretation (FCE) generation;

[0116] An explanatory chain is constructed from four dimensions to collaboratively generate human-understandable explanations for decision-making:

[0117] (1) Key viewpoint chain: The viewpoint nodes in the tracing path are arranged in descending order of cumulative impact intensity, presenting the core viewpoints with the greatest influence in the decision-making process, specifically:

[0118] Key viewpoint chain :

[0119] ;

[0120] Each element Indicates: Viewpoint Node The time of the proposal The cumulative impact strength on the conclusion The chain is based on Arranged in descending order, these represent the viewpoints that have the greatest influence in the decision-making process.

[0121] (2) Evidence Support Chain: The evidence nodes in the tracing path are arranged in descending order of reliability, presenting the evidence support system for the decision-making process, specifically as follows:

[0122] Chain of evidence :

[0123] ;

[0124] Each element Indicates: Evidence Node Source of evidence (e.g., literature citations, experimental data, industry reports, etc.) Evidence reliability The chains are arranged in descending order of reliability.

[0125] (3) Objection Processing Chain: The objection nodes in the tracing path are arranged according to their state priority, presenting the processing status of objections, specifically as follows:

[0126] Opposition to processing chain :

[0127] ;

[0128] Each element Indicates: Nodes of dissenting opinions final state Processing time (The interval from submission to processing).

[0129] (4) Expert Contribution Chain: Experts are sorted in descending order of normalized responsibility score to present the distribution of expert contributions. A bidirectional node index is constructed to record the corresponding information of each node in the traceability path subgraph in the four chains, realizing O(1) time query of cross-chain information of any node, ensuring the coordinated linkage of the four chains, specifically:

[0130] Expert contribution chain :

[0131] ;

[0132] Each element Indicates: Expert Normalized responsibility attribution score Contribution Qualitative Labels (e.g., "core contributor", "significant participant", "general participant" etc.).

[0133] S5: Output of traceability report;

[0134] Integrating all the above analytical results with corresponding quantitative indicators, a structured decision tracing and responsibility attribution report is output. The report includes: a four-chain node list, a bidirectional node index mapping table, an objection handling completeness rate indicator, an objection adoption influence index indicator, an expert contribution list sorted in descending order of responsibility score, and timeline visualization data for each chain. Specifically:

[0135] To enable collaborative querying and cross-referencing among the four chains, this invention constructs a Bidirectional Node Index (BNI):

[0136] ;

[0137] in For nodes Cross-reference records in the four chains:

[0138] ;

[0139] Based on BNI, it is possible to Complete the query to find the position of any node in any chain within a given time. Cross-chain query capability.

[0140] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for reverse tracing of constrained decision paths based on temporal argumentation graphs, characterized in that, Includes the following steps: S1: Collect the speeches and final decision conclusions of multiple experts in the review meeting, extract the viewpoints, evidence, conclusions, objections and the relationship between them through natural language processing, and construct the time sequence argumentation graph G=(V,E,T,W,Φ,Δ); The node set V contains four types of nodes: opinion nodes, evidence nodes, final conclusion nodes, and opposing nodes; the directed edge set E contains five types of edges: supporting edges, opposing edges, rebuttal edges, reference edges, and derived edges. T is the node timestamp function, used to characterize the generation time of each node; W is the edge weight function, used to characterize the strength of each argument relationship; Φ is the expert authority function, used to characterize the degree of authority of each expert; Δ is the time decay function; S2: Based on the temporal argumentation graph, perform temporal influence propagation model calculation, sequentially calculate the influence potential of each node, the temporal propagation influence of each directed edge, and the cumulative influence intensity of each node on the final conclusion node, forming an influence intensity calculation result that includes influence potential, temporal propagation influence, and cumulative influence intensity; S3: Starting from the final conclusion node, perform a triple-constraint reverse breadth-first search based on the influence intensity calculation results. During the search process, simultaneously apply strong influence constraints, causal integrity constraints, and time consistency constraints to filter out source path subgraphs that satisfy all triple constraints and the corresponding influence intensity cache. S4: Based on the source tracing path subgraph and impact intensity cache, the life cycle management analysis of objections, the expert responsibility attribution analysis, and the generation of four-chain collaborative interpretation are performed in a coordinated manner to obtain the four-chain collaborative interpretation results, which include the key viewpoint chain, the evidence support chain, the objection handling chain, and the expert contribution chain. S5: Integrate the results of the four-chain collaborative explanation with the corresponding quantitative indicators, and output a structured decision tracing and responsibility attribution report.

2. The method for reverse tracing of constrained decision paths based on temporal argumentation graphs according to claim 1, characterized in that, In S1, the edge weight function is determined based on the edge type and temporal context: the weight of a supporting edge is proportional to the confidence level of the viewpoint node it connects to and the authority of the expert who proposed the viewpoint node; the weight of an opposing edge is proportional to the strength of the argument of the opposing node it connects to; the weight of a rebuttal edge is proportional to the specificity of the rebuttal relationship it carries; the weight of a citation edge is proportional to the reliability of the evidence node it connects to; and the weight of a derived edge is proportional to the degree of logical consistency between the final conclusion node it connects to and the upstream premise node.

3. The method for reverse tracing of constrained decision paths based on temporal argumentation graphs according to claim 1, characterized in that, In S2, the formula for calculating the influence potential of any node is: ； v is any node in the node set V. To propose nodes The expert authority, confidence(v) is the confidence level inherent in node v, δ v Node type weight; The formula for calculating the temporal propagation effect of directed edges is: ； Where e is the directed edge from source node u to target node v, and W(e) is the weight of the directed edge. The time decay function adopts an exponential decay form, and the calculation formula is as follows: ； Where λ>0 is the decay rate hyperparameter, t u t is the timestamp of the source node. v The timestamp of the target node v.

4. The method for reverse tracing of constrained decision paths based on temporal argumentation graphs according to claim 3, characterized in that, In S2, the formula for calculating the cumulative influence intensity of each node on the final conclusion node is as follows: ； In this formula, S path (v,c) is the weighted sum of the temporal propagation effects of all paths from node v to the final conclusion node c, S direct (v,c) represents the maximum direct temporal propagation influence of node v on the ancestor nodes of the final conclusion node c, and α∈[0,1] is the path-direct equilibrium coefficient.

5. The method for reverse tracing of constrained decision paths based on temporal argumentation graphs according to claim 4, characterized in that, In S3, the triple constraint specifically includes: Strong influence constraint: Only nodes whose cumulative influence on the final conclusion node is not less than a preset node threshold, and directed edges whose temporal propagation influence is not less than a preset edge threshold are retained; Causal integrity constraint: The sequence of adjacent triplet node types on the reverse search path must belong to the predefined legal causal pattern library and not to the illegal causal pattern library; Time consistency constraint: The source node corresponding to each edge on the reverse search path is generated no later than the target node, and the time interval between adjacent nodes does not exceed the preset maximum allowed interval.

6. The method for reverse tracing of constrained decision paths based on temporal argumentation graphs according to claim 5, characterized in that, In S3, the reverse breadth-first search employs three optimization measures: Influence strength memoization search: The cumulative influence strength is calculated incrementally using dynamic programming, and the influence strength results of already calculated nodes are cached to avoid duplicate calculations; Causal pattern library pruning: During the search process, the current path prefix is ​​detected, and if an illegal causal pattern is matched, the search for that branch is terminated; Hierarchical Time Index: A pre-built bucket index structure based on timestamps is used to search only the buckets whose time is no later than the current node, reducing the complexity of inbound edge retrieval.

7. The method for reverse tracing of constrained decision paths based on temporal argumentation graphs according to claim 6, characterized in that, In S4, the lifecycle management analysis of objections specifically includes: The lifecycle of dissenting opinions is modeled as a finite state machine containing five states: raised, under discussion, refuted, adopted, and pending processing. The Obligation Completeness Rate (OHCR) is calculated using the following formula: ； Among them, V opposition Let state(o) be the set of all opposing nodes, and state(o) be the final state of opposing node o. The Opposition Adoption Impact Index (OAI) is calculated using the following formula: ； Where I(o,c) is the cumulative influence strength of the opposing node o on the final conclusion node c, and V opinion Let ϵ be the set of all viewpoint nodes, and let ϵ be a small constant to prevent division by zero.

8. The method for reverse tracing of constrained decision paths based on temporal argumentation graphs according to claim 7, characterized in that, In S4, expert responsibility attribution analysis quantifies the experts' contributions to the final conclusion node from four dimensions: Impact and contribution of viewpoints: ； Evidence supporting the contribution: ； Contribution of dissenting opinions: ； Time-series contribution factor: ； Among them, V evidence The set of evidence nodes, e represents the target expert, reliability(v) represents the reliability of the evidence node, and sign(o) is the polarity function of the opposing opinion, taking a value of +1 when adopted, -1 when refuted, and 0 when unresolved. The contribution values ​​of the four dimensions are weighted and aggregated to obtain the expert's comprehensive responsibility attribution score. ； in, For dimension weights, satisfying The comprehensive responsibility attribution scores were normalized to obtain the relative responsibility coefficients of each expert.

9. The method for reverse tracing of constrained decision paths based on temporal argumentation graphs according to claim 1, characterized in that, In S4, the four-chain collaborative interpretation generation process constructs a bidirectional node index, which records the ranking of each node in the key viewpoint chain, the source information in the evidence support chain, the processing status in the opposition processing chain, and the responsibility attribution information in the expert contribution chain for each node in the tracing path subgraph, so as to realize cross-chain information O(1) time query for any node.

10. The method for reverse tracing of constrained decision paths based on temporal argumentation graphs according to claim 1, characterized in that, In S5, the structured decision tracing and responsibility attribution report includes: a four-chain node list, a two-way node index mapping table, a completeness index for handling objections, an impact index for the adoption of objections, an expert contribution list sorted in descending order of responsibility score, and timeline visualization data corresponding to each chain.

Images