Industrial chain risk data early warning method and system based on digital intelligence

Abstract

The invention provides an industrial chain risk data early warning method and system based on digital intelligence, and relates to the technical field of intelligent data analys.The method comprises the steps that S1, semantic data of an industrial chain entity in an unstructured text is obtained in real time, a syntactic tree with a core predicate as a root node is constructed, and a syntactic tree is constructed; the dependency type and the dependency arc strength of the entity node are marked in the syntactic tree; calculating a semantic weight according to the hierarchical depth of the node in the syntactic tree and the dependency arc intensity, so as to determine three entity nodes with the highest causal influence as target detection points; and S2, mapping the target detection points to a two-dimensional coordinate system with a syntactic tree root node as an original point, performing normalization processing on semantic weight vectors of the three target detection points, and generating a minimum convex polygon covering all associated entities through a Delou inner triangulation algorithm. According to the method, the unstructured text is analyzed in real time to construct the precise semantic network, and identification, early warning and intelligent prevention and control of industrial chain risks are realized.

Description

Technical Field

[0001] The present invention relates to the field of intelligent data analysis technology, and in particular to a digital-based industrial chain risk data early warning method and system. Background Art

[0002] Traditional industrial chain risk data early warning methods face many challenges. Traditional methods rely on manual collection and analysis of data, and have problems with data lag and one-sidedness.

[0003] For example, when a supplier of core materials for power batteries experiences a reduction in production capacity due to equipment failure, traditional methods require obtaining information through corporate reports, manual visits, and other channels. This may result in data delays of weeks or even months before it is transmitted to other links in the industrial chain, making it difficult for manufacturers to adjust production plans in a timely manner, and easily leading to inventory backlogs or delivery delays. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a digital-based industrial chain risk data warning method and system, so as to realize accurate identification of industrial chain risks, early warning and dynamic monitoring of the entire process, and improve the timeliness of industrial chain risk prevention and control.

[0005] In order to solve the above technical problems, the technical solutions of the present invention are as follows:

[0006] In the first aspect, a digitalized industrial chain risk data early warning method is provided, the method comprising:

[0007] Step S1: Acquire semantic data of industry chain entities in unstructured text in real time, construct a syntax tree with the core predicate as the root node, and annotate the dependency type and dependency arc strength of the entity nodes in the syntax tree; calculate the semantic weight based on the hierarchical depth and dependency arc strength of the nodes in the syntax tree to determine the three entity nodes with the highest causal influence as target detection points;

[0008] Step S2: Map the target detection points to a two-dimensional coordinate system with the root node of the syntax tree as the origin, normalize the semantic weight vectors of the three target detection points, and generate a minimum convex polygon covering all associated entities using the Delaunay triangulation algorithm; calculate the intersection of the perpendicular bisectors of each side of the polygon as the geometric center, and project this center point below the common parent node of adjacent dependency paths in the syntax tree to generate a path correction point;

[0009] Step S3, extracting the predicate-argument structure of the path correction point and the adjacent nodes in the original dependency path to form a corrected dependency path;

[0010] Step S4: updating the dynamic industry knowledge network based on the revised dependency path, and performing strength fusion of the reconstructed cross-level semantic edges and the original syntactic edges to obtain the comprehensive strength of the fused path;

[0011] Step S5: Based on the comprehensive strength of the fusion path, a risk diffusion simulation is performed through the risk conduction analysis engine to obtain a risk quantification result;

[0012] Step S6: Match the risk quantification result with the preset threshold in the industry risk threshold library.

[0013] Secondly, a digital and intelligent industry chain risk data early warning system includes:

[0014] The semantic parsing module is used to extract the semantics of industry chain entities from unstructured text, construct a syntactic tree with the core predicate as the root node, annotate the dependency type and dependency arc strength, and then calculate the semantic weight to identify the final three target detection points of causal influence;

[0015] The geometric modeling module is used to map the target detection points into a two-dimensional coordinate system, normalize the semantic vector, generate the minimum convex polygon through Delaunay triangulation, calculate the geometric center and project it into the syntax tree structure, and generate the path correction point;

[0016] The path correction module is used to extract the predicate-argument structure of the path correction point and adjacent nodes, and reconstruct the corrected dependency path;

[0017] The knowledge fusion module is used to update the dynamic industry knowledge network based on the revised dependency path, fuse the connection strength of cross-level semantic edges with the original syntactic edges, and generate the comprehensive strength of the fused path;

[0018] The risk simulation module is used to simulate the risk diffusion of the industrial chain based on the comprehensive strength of the fusion path and use the risk transmission analysis engine to obtain risk quantification results;

[0019] The threshold matching module is used to compare the risk quantification results with the preset thresholds in the industry risk threshold library to automatically determine and output the risk warning level.

[0020] According to a third aspect, a computing device includes:

[0021] one or more processors;

[0022] The storage device is used to store one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement the method.

[0023] In a fourth aspect, a computer-readable storage medium stores a program, which implements the method when executed by a processor.

[0024] The above solution of the present invention includes at least the following beneficial effects:

[0025] By constructing a multi-level syntactic tree through dependency syntactic parsing and combining it with semantic association strength calculation, the entity nodes with the highest causal influence in the industrial chain can be accurately identified, avoiding the problem of missed or misjudgment of key nodes caused by semantic ambiguity in traditional methods. Based on the Delaunay triangulation algorithm, the minimum convex polygon is generated and the semantic geometric center is located. Path correction points are dynamically inserted to effectively resolve the association deviation caused by hierarchical breaks or semantic conflicts in the original dependency path, thereby improving the coherence and completeness of risk transmission path analysis. By reconstructing the strength fusion of cross-level semantic edges and original syntactic edges, the limitations of traditional single-layer semantic analysis are broken through, and the ability to mine hidden risk transmission paths in the industrial chain is enhanced. It is particularly suitable for multi-level risk association analysis in complex supply chain scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a flow chart of a digitalized industrial chain risk data early warning method provided by an embodiment of the present invention.

[0027] Figure 2 This is a schematic diagram of a digital-based industrial chain risk data early warning system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0028] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0029] like Figure 1 As shown, an embodiment of the present invention proposes a digitalized industrial chain risk data early warning method, which includes the following steps:

[0030] Step S1: Acquire semantic data of industry chain entities in unstructured text in real time, construct a syntax tree with the core predicate as the root node, and annotate the dependency type and dependency arc strength of the entity nodes in the syntax tree; calculate the semantic weight based on the hierarchical depth and dependency arc strength of the nodes in the syntax tree to determine the three entity nodes with the highest causal influence as target detection points;

[0031] Step S2: Map the target detection points to a two-dimensional coordinate system with the root node of the syntax tree as the origin, normalize the semantic weight vectors of the three target detection points, and generate a minimum convex polygon covering all associated entities using the Delaunay triangulation algorithm; calculate the intersection of the perpendicular bisectors of each side of the polygon as the geometric center, and project this center point below the common parent node of adjacent dependency paths in the syntax tree to generate a path correction point;

[0032] Step S3, extracting the predicate-argument structure of the path correction point and the adjacent nodes in the original dependency path to form a corrected dependency path;

[0033] Step S4: updating the dynamic industry knowledge network based on the revised dependency path, and performing strength fusion of the reconstructed cross-level semantic edges and the original syntactic edges to obtain the comprehensive strength of the fused path;

[0034] Step S5: Based on the comprehensive strength of the fusion path, a risk diffusion simulation is performed through the risk conduction analysis engine to obtain a risk quantification result;

[0035] Step S6: Match the risk quantification result with the preset threshold in the industry risk threshold library.

[0036] In an embodiment of the present invention, by acquiring the semantic data of the industrial chain entities in the unstructured text in real time, constructing a syntax tree and annotating relevant information, it breaks through the limitation of traditional data processing relying on structured data, and can effectively mine the risk clues hidden in the text. Calculating semantics and determining target detection points can focus on entity nodes that have a key impact on industrial chain risks, avoiding the waste of resources caused by analyzing all nodes. The target detection points are mapped to a two-dimensional coordinate system and processed, and the spatial correlation relationship between entity nodes is established through geometric methods. Compared with the traditional method of analyzing only from the text grammar level, this method more intuitively shows the semantic connection between entity nodes, generates path correction points, and can optimize the dependency path between entity nodes.

[0037] Extracting the predicate-argument structure of path correction points and adjacent nodes to form a revised dependency path accurately depicts the semantic and logical relationships between entity nodes. Compared to traditional, more coarse dependency analysis, this operation refines the connections between nodes, ensuring that the dependency path reflects both grammatical structure and semantic connotation, providing strong support for the precise representation of risk transmission paths. Updating the dynamic industrial knowledge network and integrating the strengths of semantic and syntactic edges enriches and improves the industrial chain knowledge system, breaking the drawbacks of traditional knowledge networks, which are fixed in structure and difficult to adapt to dynamic changes. The resulting comprehensive strength of the fused path comprehensively measures the closeness and importance of connections between entity nodes, enabling the industrial knowledge network to more realistically reflect the complex relationships in the actual operation of the industrial chain. A risk diffusion simulation is performed based on the comprehensive strength of the fused path. Using the constructed precise path and strength relationships, the diffusion of risk within the industrial chain is simulated. Compared to traditional, simple risk inference, this operation can more scientifically and accurately predict the scope and extent of risk diffusion, allowing for the timely development of effective response strategies to mitigate risk losses.

[0038] In a preferred embodiment of the present invention, step S1 acquires semantic data of industry chain entities in unstructured text in real time, constructs a syntax tree with a core predicate as a root node, and annotates the dependency type and dependency arc strength of the entity nodes in the syntax tree; calculates semantic weights based on the hierarchical depth and dependency arc strength of the nodes in the syntax tree to determine three entity nodes with the highest causal influence as target detection points, which may include:

[0039] Step S110 , performing dependency syntax parsing on the unstructured text, identifying the core predicates and corresponding argument roles between industry chain entities, and constructing a multi-level syntax tree with the core predicate as the root node based on the predicate-argument relationship;

[0040] Step S111: traverse all entity nodes in the syntax tree, mark the dependency type and dependency arc strength for each node, and dynamically assign an initial strength value to the dependency arc strength according to a preset syntax relationship type mapping table;

[0041] Step S112: Adjust the influence of dependency strength according to the hierarchical depth of the entity node in the syntax tree. Starting from the root node hierarchical depth, the influence of dependency strength gradually decreases with each layer extending toward the leaf node. The semantic association strength of the node is calculated by combining the influence of hierarchical depth on dependency strength with the dependency arc strength.

[0042] In step S113, all entity nodes in the syntactic tree are sorted based on the semantic association strength, the three nodes with the final association strength are determined as target detection points, and the dependency type and hierarchical position are recorded. The target detection points are used to represent the core entities of causal association in the text.

[0043] In an embodiment of the present invention, for unstructured texts such as news reports, corporate announcements, and policy documents, a dependency parsing method in natural language processing technology is used. During the parsing process, words that can express core actions, states, or relationships are found from the sentences in the text and determined as core predicates. For example, in the sentence "A lithium battery manufacturer has suspended some production lines due to rising raw material prices," "suspend" is the core predicate. Entities with semantic relationships with the core predicate are identified around the core predicate. These entities are called arguments in linguistics, and the role played by each argument is clarified. In the above sentence, "lithium battery manufacturer" is the executor of the action and plays the role of "agent"; "part of the production line" is the object of the action and plays the role of "recipient"; "rising raw material prices" explains the reason for the action and plays the role of "cause." Taking the core predicate as the root node, based on the semantic relationship between the arguments and the core predicate, and in accordance with the subject-verb-object, attributive-adverbial-complement grammatical rules, all arguments and the core predicate are constructed into a multi-level syntactic tree structure, just like a big tree, with the core predicate as the root node and the arguments as branches and leaves, which are connected to each other through grammatical relationships to form a clear hierarchical relationship.

[0044] After the syntax tree is constructed, all entity nodes in the syntax tree are traversed one by one. During the traversal, the nodes are labeled with dependency types based on their grammatical relationship with the core predicate, referring to a preset syntax relationship type mapping table. For example, if the node and the core predicate have a subject-predicate relationship, it is labeled with a "subject" dependency type; if it has a verb-object relationship, it is labeled with an "object" dependency type. Based on the strength standards corresponding to different dependency types in the syntax relationship type mapping table, the dependency arcs connecting each node are dynamically assigned initial strength values. The mapping table stipulates that key relationships that directly affect the core action, such as "agent" and "patient," have high initial strength values, while relationships that provide auxiliary explanations of time and location have low initial strength values. For example, the initial strength value of the dependency arc between "lithium battery manufacturer" (agent) and "pause" might be set to 80 (assuming a strength range of 0-100), the initial strength value of the dependency arc between "partial production line" (patient) and "pause" might be set to 75, and the initial strength value of the dependency arc between "raw material price increase" (reason) and "pause" might be set to 60.

[0045] The root node (core predicate) of the syntactic tree is defined as the starting level, or level 1. Starting from the root node, the level depth increases by 1 for each level toward the leaf nodes. For example, the node directly connected to the root node is level 2, the node connected to the next level node is level 3, and so on. Because nodes farther from the root node have less influence on the core relationship expressed by the core predicate, a pre-set rule is used to gradually reduce the influence of dependency strength as level depth increases. For example, for each additional level, the influence of dependency strength decreases by 10%. The dependency strength adjusted for level depth is then combined with the node's original dependency arc strength to calculate the final semantic association strength for each node. For example, suppose the node "Raw material prices rise" originally had a dependency arc strength of 60 and was located at level 3. After level depth adjustment, its dependency strength becomes 60 × (1 - 10% × 2) = 48. This 48 represents the semantic association strength of the node after taking into account both the level depth and the initial dependency arc strength. After obtaining the semantic association strengths for all entity nodes, the syntactic tree is sorted from highest to lowest strength. After sorting, the three nodes with the highest semantic association strengths are selected as target checkpoints. The dependency types of these three target checkpoints and their hierarchical positions in the syntactic tree are recorded in detail. These target checkpoints can most directly and critically reflect the causal relationships inherent in the text.

[0046] By converting unstructured text into a structured, multi-level syntactic tree, the method transforms complex, unorganized text into a clear, organized entity-relationship network, providing a structured data foundation for risk early warning and addressing the difficulty traditional methods have in processing unstructured text. By labeling entity nodes with dependency types and assigning initial strength values, the method quantifies the relationships between entities. This allows for rapid differentiation between key and secondary entities within a vast amount of industry chain information, reducing the interference of invalid information and avoiding risk misjudgments caused by information overload. The method adjusts dependency strength based on the node's hierarchical depth in the syntactic tree, fully accounting for the proximity and importance of semantic relationships, ensuring that the calculated semantic association strength is more consistent with actual semantic logic. This approach avoids potential misjudgments that can arise from relying solely on initial strength values, more accurately identifying entity nodes with significant influence on industry chain risks, and ultimately selecting three target detection points for semantic association strength from a large number of entity nodes. This allows for a focused analysis of the core entities with the ultimate causal influence on industry chain risks, enabling rapid identification of potential risk sources within a short period of time.

[0047] In a preferred embodiment of the present invention, step S2, mapping the target detection points to a two-dimensional coordinate system with the root node of the syntax tree as the origin, normalizing the semantic weight vectors of the three target detection points, generating a minimum convex polygon covering all associated entities using the Delaunay triangulation algorithm; calculating the intersection of the perpendicular bisectors of each side of the polygon as the geometric center, projecting the center point below the common parent node of adjacent dependency paths in the syntax tree, and generating a path correction point, may include:

[0048] Step S220 , mapping the hierarchical position and dependency type of the target detection point to a two-dimensional coordinate system with the root node of the syntax tree as the origin, where the abscissa represents the dependency path number of the node's horizontal branch in the syntax tree, and the ordinate represents the hierarchical depth of the node relative to the root node;

[0049] Step S221 , normalizing the coordinate values ​​based on the semantic association strength of the target detection point so that the coordinate values ​​of nodes at different levels and branches are unified into a preset numerical range;

[0050] Step S222: Based on the normalized node coordinates, the Delaunay triangulation algorithm is used to connect the target detection point and the associated entity nodes in the adjacent dependency path to generate a minimum convex polygon covering all associated entities, and the edges of the minimum convex polygon represent the potential semantic association paths between entities;

[0051] Step S223: Calculate the intersection of the perpendicular bisectors of the sides of the minimum convex polygon as the semantic geometric center point, and reversely map the coordinates of the center point into the syntax tree structure, locating it below the nearest common parent node of the dependency path where the target detection point is located;

[0052] In step S224 , a new node is inserted between the common parent node and the direct child node as a path correction point, and the dependency arc strength is dynamically adjusted according to the semantic association strength between the geometric center point and the original path node to correct the semantic coherence of the original dependency path.

[0053] In this embodiment of the present invention, after obtaining the three target detection points determined in step S1, the hierarchical position and dependency type of each target detection point in the syntax tree are first sorted out, with the root node of the syntax tree serving as the origin of the two-dimensional coordinate system, i.e., position (0, 0). The abscissa is determined by sequentially numbering the dependency paths of the horizontal branches in the syntax tree from left to right, with the sequence number of the dependency path where the target detection point is located serving as the abscissa value. The ordinate is determined by the hierarchical depth of the target detection point relative to the root node, with the root node at level 1. The hierarchical depth increases by 1 for each downward step, and this depth value serves as the ordinate value. For example, in the syntax tree of "A car manufacturer adjusts its production plan due to a chip shortage", if "chip shortage" (cause), "car manufacturer" (agent), and "production plan" (respondent) are the three target detection points, "chip shortage" is in the second layer and the second horizontal dependency path below the root node "adjustment", and its coordinates are recorded as (2, 2); "car manufacturer" is in the first layer and the first horizontal dependency path, with coordinates (1, 1); "production plan" is also in the first layer and the first horizontal dependency path, with coordinates (1, 1). Due to the different hierarchical depths and dependency paths of different target detection points, their coordinate value ranges vary greatly. In order to facilitate analysis and comparison, the coordinate values ​​need to be normalized. The specific operation is to first set a preset numerical interval, such as [0, 1]. Then, according to the semantic association strength of each target detection point, its horizontal and vertical coordinate values ​​are scaled and translated. Nodes with higher semantic association strength will try to make their coordinate values ​​close to the upper limit of the preset interval during the normalization process; nodes with lower strength will have coordinate values ​​closer to the lower limit. In this way, the coordinate values ​​of all target detection points are unified into a preset numerical range, eliminating the incomparability of coordinate values ​​caused by hierarchical and path differences.

[0054] After normalizing the coordinates of the target detection points, the Delaunay triangulation algorithm is used. This algorithm uses these target detection points as a basis, while also considering the associated entity nodes in their adjacent dependency paths. It then searches for a final connection in the two-dimensional coordinate system, connecting these nodes to form a minimum convex polygon that covers all associated entities. Each edge of this polygon represents a potential semantic association path between the entities. Once the minimum convex polygon is obtained, the perpendicular bisector of each edge of the polygon is calculated. The perpendicular bisector is the line that perpendicularly bisects the polygon's edges. The intersection of all perpendicular bisectors is found. This intersection is the semantic geometric center, which geometrically represents the "center" of all associated entities. The coordinates of this center in the two-dimensional coordinate system are then reverse-mapped back to the syntax tree structure. Specifically, the center is located below the nearest common parent node of the dependency path containing the target detection point. Specifically, within the syntax tree hierarchy, a node that both covers the target detection point and is a common parent node is found, and the center is mapped below this node. A new node is inserted between the identified common parent node and its direct child nodes. This new node serves as the path correction point. After insertion, the dependency arc strength is dynamically adjusted based on the semantic association strength between the semantic geometric center point and the original path nodes. If the center point has a strong semantic association with an original node, the dependency arc strength connecting them is increased; otherwise, it is decreased. In this way, the semantic coherence of the original dependency path is corrected, making the semantic relationships between entity nodes in the syntactic tree more consistent with actual logical connections.

[0055] Target detection points are mapped to a two-dimensional coordinate system. A coordinate system is constructed using hierarchical positions and dependency paths, transforming abstract semantic relationships in the syntax tree into intuitive spatial positional relationships, reducing the complexity of semantic relationship analysis. Normalization eliminates coordinate value deviations caused by differences in node hierarchy and paths, making node data with different attributes comparable. This ensures that each node participates in subsequent algorithmic processing based on a unified standard, preventing the impact of varying data scales on the accuracy of analysis results and improving the scientific nature of data processing. The Delaunay triangulation algorithm is used to generate minimum convex polygons, visually representing the potential semantic connection paths between entities. Compared to traditional text analysis, this representation can quickly help analysts identify hidden connections between entities, providing a more efficient analytical perspective for exploring risk transmission paths within the industrial chain. Determining the semantic geometric center point and mapping it back to the syntax tree provides a clear reference point for correcting dependency paths. By mapping the geometric center point to the hierarchical structure of the syntax tree, the semantic core of the associated entities can be grasped holistically, providing a basis for accurately locating path corrections and enhancing the integrity and logic of semantic relationship analysis. Inserting path correction points and dynamically adjusting the dependency arc strength effectively corrects possible semantic faults or logical irrationalities in the original dependency path, thereby improving the semantic coherence between entities in the syntax tree and ensuring that the risk transmission analysis of the industrial chain based on this path is more in line with actual conditions.

[0056] In a preferred embodiment of the present invention, step S3 of extracting the predicate-argument structure of the path correction point and the adjacent nodes in the original dependency path to form a corrected dependency path may include:

[0057] Step S330 , based on the position of the path correction point and the inherited dependency type, locate the connection relationship with the parent node and child nodes in the original dependency path, and extract the predicate-argument structure of the correction point and the upstream and downstream nodes;

[0058] Step S331: Check whether there is a semantic conflict between the dependency types of the correction point and its parent node and child node. If there is a logical contradiction between the dependency type from the parent node to the correction point and the dependency type from the correction point to the child node, reallocate the dependency arc direction according to the preset dependency type rule, and retain the dependency type from the parent node to the correction point.

[0059] Step S332 , delete the redundant connection nodes in the original dependency path that are broken due to the insertion of the correction point, and re-establish the dependency relationship between the correction point and the adjusted parent node and child nodes to form a cross-level corrected dependency path.

[0060] In an embodiment of the present invention, based on the specific location of a path correction point in the syntax tree and the dependency type it inherits from its parent node, the connection relationship between the correction point and the parent node and child node in the original dependency path is accurately found. For example, in the syntax tree of "A certain electronics company delayed order delivery due to raw material price increases," if "raw material price increases" is a path correction point, its dependency type is "cause," its parent node is "delay" (core predicate), and its child node is "order delivery." Then, the predicate-argument structure between the correction point and the upstream and downstream nodes (i.e., parent node and child node) is extracted. For "raw material price increases" and "delay," "raw material price increases" serves as the "cause" argument, and "delay" serves as the core predicate. Although there is no direct predicate-argument relationship between "raw material price increases" and "order delivery," the insertion of the correction point requires that they be incorporated into the new analysis structure to clarify their semantic and logical connection. In this way, the local predicate-argument structure relationship containing the correction point is sorted out. After extracting the predicate-argument structure, carefully check whether there are any semantic conflicts between the dependency types of the correction point and the parent and child nodes. For example, if the dependency type from the parent node "Adjusting Production Plan" to the correction point "Declining Market Demand" is labeled "Cause," while the dependency type from the correction point "Declining Market Demand" to the child node "Reducing Production" is labeled "Result," this is logically reasonable. However, if the dependency type from the parent node to the correction point is labeled "Result," and the dependency type from the correction point to the child node is also labeled "Result," there is a logical contradiction, which is a semantic conflict.

[0061] When a semantic conflict is discovered, dependency arc directions are reassigned according to pre-defined dependency type rules. These rules explicitly prioritize preserving the dependency type from the parent node to the correction point. For example, if a conflict occurs, the dependency type from the correction point to the child node is adjusted to the same "cause" as the one from the parent node to the correction point. This ensures semantic consistency across the entire dependency relationship and avoids inconsistencies and confusion. The insertion of a path correction point may introduce unnecessary redundant nodes in the original dependency path, disrupting the coherence and accuracy of the revised dependency path. Therefore, the original dependency path needs to be cleaned up and these redundant nodes removed. Next, the dependency relationship between the path correction point and the adjusted parent and child nodes is reestablished. For example, after removing the redundant nodes, the dependency connections between "Raw Material Price Increase" (correction point) and "Delay" (parent node) and "Order Delivery" (child node) are clarified, thus constructing a revised dependency path across different hierarchies. This new dependency path not only incorporates the semantic additions provided by the correction point but also incorporates the valid information from the original path, forming a complete, coherent, and semantically logical dependency chain.

[0062] By precisely locating the connections between path correction points and original path nodes and extracting predicate-argument structures, the correction points are naturally integrated into the original semantic relationship network, avoiding disruption to the original logical system caused by the insertion of correction points. This lays the foundation for constructing a more complete dependency path and ensures the integrity and accuracy of semantic information. By detecting and correcting semantic conflicts in dependency types, potential logical contradictions in the syntactic tree are eliminated, ensuring the consistency and rationality of dependency relationships at the semantic level, avoiding misjudgments of industrial chain risk transmission paths due to incorrect semantic associations, and improving the reliability of risk warning analysis. By removing redundant nodes and re-establishing dependency relationships, the original dependency path is streamlined and optimized, making the revised dependency path more concise, clear, and logically coherent. The cross-level revised dependency path can more realistically reflect the semantic connections between industrial chain entities, providing an efficient and accurate analysis path for subsequent path-based risk analysis and warning, and enhancing the effectiveness of risk warnings.

[0063] In a preferred embodiment of the present invention, the above step S4, updating the dynamic industry knowledge network based on the modified dependency path, and performing strength fusion of the reconstructed cross-level semantic edges with the original syntactic edges to obtain the comprehensive strength of the fused path, may include:

[0064] Step S440 , based on the revised dependency path and dependency type and strength value, a cross-level semantic edge is added to the dynamic industry semantic network, and associated and integrated with the original syntactic edge;

[0065] Step S441: If the dependency type of the cross-level semantic edge between the same nodes is consistent with that of the original syntactic edge, they are merged into a single semantic edge, and the strength value is dynamically adjusted according to the dependency strength of the original path and the revised path;

[0066] Step S442: If there is a conflict between the dependency types of the cross-level semantic edges and the original syntactic edges between the same nodes, the semantic edges associated with the revised path are retained according to the dependency type rule, and the association strength is recalculated based on the level depth and dependency strength.

[0067] In step S443 , the integrated semantic edges and association strengths are used as quantitative indicators of the fusion path and are synchronously updated to the dynamic semantic network to obtain the comprehensive strength of the fusion path.

[0068] In an embodiment of the present invention, based on the revised dependency path formed in step S3, the dependency type (such as "cause" and "agent") and the corresponding dependency strength value (such as the value of the initial strength value after hierarchical adjustment) of each node in the path are extracted. Then, in the dynamic industry semantic network (i.e., the pre-constructed industry chain entity relationship network), new semantic edges are created for the newly added cross-level connections in the revised path (such as the connection established through the correction point between upstream and downstream nodes that were not directly related). These semantic edges represent the newly identified semantic associations between entities. At the same time, the new semantic edges are associated with the syntactic edges that already exist in the original syntactic tree (i.e., the original dependency relationship constructed in step S1). If the revised dependency path adds a cross-level association between "battery raw material price fluctuations" and "vehicle delivery cycle", a semantic edge from "battery raw material supplier" to "vehicle manufacturer" is added to the semantic network, and the dependency type is marked as "indirect impact" and the strength value is the adjusted value. Traverse all semantic edges between the same entity node pair in the dynamic semantic network and check whether the dependency type of the cross-level semantic edge is consistent with that of the original syntactic edge. If they match (e.g., both are "supply relationships"), the two edges are merged into a single semantic edge to avoid duplicate records. During the merge, the strength value is dynamically adjusted through regularization. For example, the average of the original path strength (e.g., 70) and the revised path strength (e.g., 65) is used as the new strength value (67.5). If the dependency type of "chip supplier" and "vehicle system manufacturer" in the original syntactic edge is "direct supply," the strength value is 70; the strength value of the revised cross-level semantic edge is 65 due to the adjusted hierarchical depth, and the dependency type is also "direct supply," then the strength value after the merge is adjusted to 67.5 and marked as "optimized supply relationship."

[0069] If the dependency type of the cross-level semantic edge for the same node pair conflicts with that of the original syntactic edge (e.g., the original edge is a "cooperative relationship" and the cross-level edge is a "competitive relationship"), then, based on the preset dependency type rule (e.g., "corrected path first" or "hierarchical depth first"), the semantic edge associated with the corrected path is retained, and the original conflicting edge is deleted and marked as "pending verification." After retention, the association strength is recalculated: a new strength value is generated using a regularized formula, combining the node's hierarchical depth and dependency strength in the syntactic tree. All integrated semantic edges and their associated strength values ​​(including the merged single strength value and the new strength value after conflict resolution) are used as quantitative indicators of the fused path and are synchronously updated to the dynamic semantic network. The updated network not only includes the results of the original syntactic analysis but also incorporates the semantic optimization brought about by the corrected path, forming a comprehensive strength value for each path (e.g., a value in the range of 0-100, with higher values ​​indicating closer associations between entities and greater potential for risk transmission).

[0070] By adding cross-level semantic edges and associating them with the original syntactic edges, the coverage of the industrial knowledge network is dynamically expanded. This addresses potential connections (such as cross-link indirect impacts) that traditional syntactic analysis fails to identify due to hierarchical limitations. This allows the network to more comprehensively reflect the complex relationships between entities in the industry chain, enhancing the integrity of risk analysis. By merging semantic edges of the same type and dynamically adjusting their strength, the network avoids the interference of redundant relationships. Furthermore, by integrating original and revised data, the strength of the association more objectively reflects the true connections between entities, enhancing the credibility of risk transmission analysis. A conflict resolution mechanism ensures the logical consistency of the knowledge network, prioritizing semantic edges along revised paths (which are generally more consistent with actual semantic logic) and recalculating strength based on hierarchical depth. This allows the network to adapt to dynamic changes in the industry chain (such as adjustments to business relationships) and reduces the risk misjudgment caused by static data. The updated dynamic semantic network integrates multiple sources of data (original syntactic analysis and revised path optimization) to form a unified comprehensive strength indicator, making the analysis of risk transmission paths more scientific and valuable for decision-making.

[0071] In a preferred embodiment of the present invention, the above step S5, based on the comprehensive strength of the fusion path, performs risk diffusion simulation by the risk conduction analysis engine to obtain risk quantification results, which may include:

[0072] Step S550 , based on the comprehensive strength and dependency type of the fusion paths in the dynamic industry semantic network, a multi-path risk diffusion analysis is performed through the risk transmission analysis engine;

[0073] Step S551: defining the risk transmission direction constraint based on the dependency type of the fusion path, and converting the comprehensive strength of the path into a transmission probability;

[0074] Step S552: Starting from the target detection point, analyze the risk diffusion path along the cross-level semantic edges and original syntactic edges. When the risk is transmitted to the branch node, perform multi-path diversion calculation based on the transmission probability.

[0075] Step S553: ​​During the diffusion process, dynamically calculate the cumulative risk intensity of each path and make adjustments based on the attenuation relationship between the path transmission probability and the node level depth;

[0076] In step S554, the real-time risk value of the node is quantified based on the cumulative intensity and transmission rate of the risk reaching the node to obtain the risk quantification result, including the risk value distribution of each node, the key transmission path identification and the cross-level diffusion rate.

[0077] In this embodiment of the present invention, the combined strengths of all fusion paths and their corresponding dependency types are obtained in a dynamic industry semantic network. The combined strength of the fusion path between "chip shortage" (target detection point) and "mobile phone production delay" is 70, and the dependency type is "causal relationship." The combined strength of the fusion path between "raw material price increase" and "product cost increase" is 65, and the dependency type is "cause." Based on this information, a risk transmission analysis engine is activated to perform a multi-path risk diffusion analysis on each fusion path. The engine identifies all possible risk diffusion paths starting from each target detection point and connected by different semantic edges (including cross-level semantic edges and original syntactic edges). For example, starting from "chip shortage," it may affect "mobile phone assembly plant" through the semantic edge of "supply relationship," and then affect "mobile phone shipment volume" through "production relationship," thus forming a risk diffusion path. The directional constraints of risk transmission are defined based on the dependency type of the fusion path. For example, if the dependency type is "causal relationship," risk is transmitted from the cause node to the result node; if the dependency type is "supply relationship," risk is transmitted from the supplier node to the demand node. Next, the combined strength of the fused path is converted into a transmission probability. A conversion rule can be set: for example, if the combined strength is between 0 and 30, the transmission probability is 0.1; between 31 and 60, the transmission probability is 0.3; and between 61 and 100, the transmission probability is 0.5. For example, if the combined strength is 70, the transmission probability is 0.5.

[0078] Starting from the target detection point, the risk diffusion path is analyzed step by step along cross-level semantic edges and original syntactic edges. When a risk propagates to a branch node (i.e., a node with multiple subsequent connected nodes), multi-path diversion calculations are performed based on the transmission probability of that path. For example, the target detection point "chip shortage" is connected to the two branch nodes "mobile phone assembly plant" and "tablet computer assembly plant" via the semantic edge of "supply relationship." Assuming the transmission probability of the "chip shortage to mobile phone assembly plant" path is 0.5 and the transmission probability of the "chip shortage to tablet computer assembly plant" path is 0.3, then the risk transmitted from "chip shortage" to "mobile phone assembly plant" is 0.5 of the total risk, and the risk transmitted to "tablet computer assembly plant" is 0.3 of the total risk. The remaining 0.2 of the risk may be transmitted along other paths or may not yet be transmitted. During the risk diffusion process, the cumulative risk strength of each path is dynamically calculated in real time. The cumulative risk strength is the sum of the strengths of all paths from the starting node (target detection point) to the current node. For example, the path from "chip shortage" to "mobile phone production delay" passes through the "mobile phone assembly plant." Assuming the path strength of "chip shortage → mobile phone assembly plant" is 30 and the path strength of "mobile phone assembly plant → mobile phone production delay" is 40, the cumulative risk intensity of this path is 30 + 40 = 70. The cumulative risk intensity is also adjusted based on the attenuation relationship between the path transmission probability and the node depth.

[0079] The real-time risk value of each node is quantified based on the cumulative intensity and transmission rate of the risk reaching each node. The transmission rate can be set based on actual conditions, such as the distance the risk propagates along the path per time unit (e.g., daily or weekly). Assuming a risk transmission rate of 10 intensity units per day, and the "mobile phone production delay" node arrives after three days, with an adjusted cumulative risk intensity of 44.8, the real-time risk value of this node is 44.8 + 10 × 3 = 74.8. This calculation ultimately yields risk quantification results, including the risk value distribution for each node (e.g., the risk value for "mobile phone production delay" is 74.8, the risk value for "tablet computer production delay" is 50.2, etc.), the identification of critical transmission paths (e.g., "chip shortage → mobile phone assembly plant → mobile phone production delay" is a critical path), and the cross-level diffusion rate (e.g., the cross-level diffusion rate from "chip shortage" to "mobile phone production delay" is 10 intensity units per day).

[0080] By comprehensively analyzing the fusion paths within the dynamic industrial semantic network and comprehensively considering the various complex relationships within the industrial chain, this approach can more accurately identify potential risk diffusion paths, avoid missing important risk transmission links, and provide a more complete perspective for risk assessment. By converting the comprehensive intensity into transmission probability and determining transmission direction constraints based on dependency types, the risk transmission simulation more closely aligns with realistic logical relationships. This quantitative and targeted approach improves the scientific nature and accuracy of risk simulation and enables more reasonable predictions of the direction and likelihood of risk transmission within the industrial chain. Multi-path flow calculations at branch nodes fully account for the diverse transmission modes of risk within the industrial chain. This avoids the limitations of considering only a single transmission path and more realistically reflects the spread of risk within a complex network. Dynamically calculating the cumulative risk intensity and adjusting it based on attenuation relationships account for the dynamic changes in risk during transmission and the impact of hierarchical depth on risk. This refined calculation method makes risk quantification results more realistic and enables a more accurate assessment of the risk level faced by each node. By quantifying the real-time risk value of each node, comprehensive risk quantification results are obtained, including the risk value distribution of each node, the identification of key transmission paths, and the cross-hierarchical diffusion rate.

[0081] In a preferred embodiment of the present invention, the above step S6, matching the risk quantification result with a preset threshold in the industry risk threshold library, may include:

[0082] Step S660: extracting corresponding risk threshold intervals, path transmission rate thresholds, and critical path impact thresholds from a threshold library based on the industry classification and risk type of the industrial chain node;

[0083] Step S661: If the real-time risk value of the node is greater than the risk threshold of the industry to which it belongs, it is marked as a high-risk node and an industry chain-level risk warning signal is generated;

[0084] Step S662: Construct an early warning report based on the industry chain-level risk early warning signal, marking the location of high-risk nodes, risk value deviation amplitude, and associated transmission paths, and marking the diffusion rate and path impact range that are greater than the threshold;

[0085] Step S663: Correlate and analyze the risk quantification results with the early warning report, and automatically adjust the threshold intervals and impact determination rules corresponding to the industry classification in the threshold library.

[0086] In an embodiment of the present invention, based on the risk quantification results obtained in step S5, each node in the industrial chain is classified and identified to determine its industry (such as electronics, automobiles, chemicals, etc.) and risk type (such as supply disruption, price fluctuation, policy risk, etc.). Then, from the pre-built industry risk threshold library, three types of thresholds corresponding to the industry classification and risk type of the node are extracted: risk threshold interval (such as the supply disruption risk threshold of the electronics industry is [40, 60]), path conduction rate threshold (such as daily risk diffusion intensity > 20 units is considered abnormal), and critical path impact threshold (such as a path risk cumulative intensity > 50 is considered a critical path). For example, if the "chip shortage" node belongs to the supply disruption risk type of the electronics industry, all threshold parameters related to the supply disruption of the electronics industry are extracted. The real-time risk value of each node (the quantification result of step S554) is compared with the risk threshold interval of the industry to which it belongs. If the real-time risk value of a node is greater than the upper limit of the interval (such as the upper limit of the supply disruption risk threshold of the electronics industry is 60, and the real-time risk value of the "chip shortage" node is 75), the node is marked as a high-risk node. Once a high-risk node is discovered, an industry chain-level risk warning signal is automatically generated, indicating that the industry chain has major risk hazards under the current risk scenario and requires immediate attention. Based on the generated industry chain-level risk warning signal, an early warning report is automatically constructed. The report content includes: the specific location of the high-risk node in the industry chain (such as the level it is in, the associated nodes), the risk value deviation amplitude (that is, the difference between the real-time risk value and the upper threshold limit), and the associated transmission path (such as "chip shortages causing delays in mobile phone production at mobile phone assembly plants"). At the same time, the report will also identify the diffusion rate that is greater than the threshold (for example, the actual transmission rate of a path is 25, which is greater than the threshold of 20) and the scope of the path impact (for example, the path affects 3 first-tier suppliers and 5 second-tier suppliers). The risk quantification results are correlated with the early warning report to evaluate the rationality of the threshold intervals and impact judgment rules corresponding to the industry classification in the current threshold library. For example, if the threshold interval [40, 60] fails to effectively warn of actual high-risk events in the electronics industry supply chain, the threshold interval will be automatically adjusted to [35, 55]. At the same time, based on the actual risk transmission situation, the impact judgment rules will be dynamically optimized (such as adjusting the critical path impact threshold of a certain type of risk from 50 to 45) to enable the threshold library to continuously adapt to changes in industry chain risks.

[0087] By accurately matching industry classifications and risk types to extract threshold parameters, we ensure the pertinence and applicability of risk assessment standards and improve the accuracy of risk warnings. By comparing risk values ​​with threshold intervals in real time, we can quickly identify high-risk nodes and generate industry chain-level warnings. Structured warning reports provide detailed information about risks (location, deviation magnitude, transmission path, etc.), helping enterprise managers quickly locate the source of risks, clarify the scope of impact, and develop precise response strategies. By dynamically adjusting the threshold library through correlation analysis, risk assessment standards can adapt to dynamic changes in the industry chain (such as market fluctuations and policy adjustments), continuously maintaining the effectiveness and reliability of the warning system and reducing the risk of missed or false alarms caused by static thresholds.

[0088] like Figure 2 As shown, an embodiment of the present invention further provides a digital-based industrial chain risk data early warning system, comprising:

[0089] The semantic parsing module is used to extract the semantics of industry chain entities from unstructured text, construct a syntactic tree with the core predicate as the root node, annotate the dependency type and dependency arc strength, and then calculate the semantic weight to identify the final three target detection points of causal influence;

[0090] The geometric modeling module is used to map the target detection points into a two-dimensional coordinate system, normalize the semantic vector, generate the minimum convex polygon through Delaunay triangulation, calculate the geometric center and project it into the syntax tree structure, and generate the path correction point;

[0091] The path correction module is used to extract the predicate-argument structure of the path correction point and adjacent nodes, and reconstruct the corrected dependency path;

[0092] The knowledge fusion module is used to update the dynamic industry knowledge network based on the revised dependency path, fuse the connection strength of cross-level semantic edges with the original syntactic edges, and generate the comprehensive strength of the fused path;

[0093] The risk simulation module is used to simulate the risk diffusion of the industrial chain based on the comprehensive strength of the fusion path and use the risk transmission analysis engine to obtain risk quantification results;

[0094] The threshold matching module is used to compare the risk quantification results with the preset thresholds in the industry risk threshold library to automatically determine and output the risk warning level.

[0095] It should be noted that this system is a system corresponding to the above method, and all implementation methods in the above method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0096] An embodiment of the present invention further provides a computing device comprising: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the above-described method. All implementations in the above-described method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0097] The embodiment of the present invention further provides a computer-readable storage medium storing instructions, which, when executed on a computer, causes the computer to execute the above-described method. All implementations in the above-described method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0098] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A digital-based early warning method for industrial chain risk data, characterized by: The method comprises: Step S1: Acquire semantic data of industry chain entities in unstructured text in real time, construct a syntax tree with the core predicate as the root node, and annotate the dependency type and dependency arc strength of the entity nodes in the syntax tree; calculate the semantic weight based on the hierarchical depth and dependency arc strength of the nodes in the syntax tree to determine the three entity nodes with the highest causal influence as target detection points; Step S2: Map the target detection points to a two-dimensional coordinate system with the root node of the syntax tree as the origin, normalize the semantic weight vectors of the three target detection points, and generate a minimum convex polygon covering all associated entities using the Delaunay triangulation algorithm; calculate the intersection of the perpendicular bisectors of each side of the polygon as the geometric center, and project this center point below the common parent node of adjacent dependency paths in the syntax tree to generate a path correction point; Step S3, extracting the predicate-argument structure of the path correction point and the adjacent nodes in the original dependency path to form a corrected dependency path; Step S4: updating the dynamic industry knowledge network based on the revised dependency path, and performing strength fusion of the reconstructed cross-level semantic edges and the original syntactic edges to obtain the comprehensive strength of the fused path; Step S5: Based on the comprehensive strength of the fusion path, a risk diffusion simulation is performed through the risk conduction analysis engine to obtain a risk quantification result; Step S6: Match the risk quantification result with the preset threshold in the industry risk threshold library.

2. The digitalized industrial chain risk data early warning method according to claim 1 is characterized in that: Acquire semantic data of industry chain entities in unstructured text in real time, construct a syntax tree with core predicates as root nodes, and annotate the dependency types and dependency arc strengths of entity nodes in the syntax tree; The semantic weight is calculated based on the node's hierarchical depth and dependency arc strength in the syntactic tree to determine the three entity nodes with the highest causal influence as target detection points, including: Perform dependency syntax analysis on unstructured text to identify the core predicates and corresponding argument roles between industry chain entities, and construct a multi-level syntax tree with the core predicate as the root node based on the predicate-argument relationship; Traverse all entity nodes in the syntax tree, mark the dependency type and dependency arc strength for each node, and dynamically assign the initial strength value of the dependency arc strength according to the preset syntax relationship type mapping table; The influence of dependency strength is adjusted according to the hierarchical depth of the entity node in the syntax tree. Starting from the root node hierarchical depth, the influence of dependency strength gradually decreases with each layer extending to the leaf node. The semantic association strength of the node is calculated by combining the influence of hierarchical depth on dependency strength with the dependency arc strength. All entity nodes in the syntactic tree are sorted based on the semantic association strength, and the three nodes with the final association strength are determined as target detection points. The dependency type and hierarchical position are recorded, and the target detection points are used to represent the core entities of causal relationships in the text.

3. The digitalized industrial chain risk data early warning method according to claim 2 is characterized in that: The target detection points are mapped to a two-dimensional coordinate system with the root node of the syntax tree as the origin. The semantic weight vectors of the three target detection points are normalized, and the minimum convex polygon covering all associated entities is generated using the Delaunay triangulation algorithm. Calculate the intersection of the perpendicular bisectors of each polygon side as the geometric center, project the center point below the common parent node of the adjacent dependency paths in the syntax tree, and generate the path correction point, including: Map the hierarchical position and dependency type of the target detection point to a two-dimensional coordinate system with the root node of the syntax tree as the origin, where the horizontal axis represents the dependency path number of the node in the horizontal branch of the syntax tree, and the vertical axis represents the hierarchical depth of the node relative to the root node; Normalize the coordinate values ​​based on the semantic association strength of the target detection points, so that the coordinate values ​​of nodes at different levels and branches are unified to the preset value range; According to the normalized node coordinates, the Delaunay triangulation algorithm is used to connect the target detection point and the associated entity nodes in the adjacent dependency path to generate a minimum convex polygon covering all associated entities. The edges of the minimum convex polygon represent the potential semantic association paths between entities. Calculate the intersection of the perpendicular bisectors of the sides of the minimum convex polygon as the semantic geometric center point, and reversely map the coordinates of the center point to the syntax tree structure, locating it below the nearest common parent node of the dependency path where the target detection point is located; New nodes are inserted between the common parent node and the direct child node as path correction points, and the dependency arc strength is dynamically adjusted according to the semantic association strength between the geometric center point and the original path node to correct the semantic coherence of the original dependency path.

4. The digitalized industrial chain risk data early warning method according to claim 3 is characterized in that: Extract the predicate-argument structure of the path correction point and the adjacent nodes in the original dependency path to form a corrected dependency path, including: Based on the position of the path correction point and the inherited dependency type, the connection relationship with the parent node and child node in the original dependency path is located, and the predicate-argument structure of the correction point and the upstream and downstream nodes is extracted; Check whether there is a semantic conflict between the dependency types of the correction point and the parent node and child node. If there is a logical contradiction between the dependency type from the parent node to the correction point and the dependency type from the correction point to the child node, reallocate the dependency arc direction according to the preset dependency type rules and retain the dependency type from the parent node to the correction point. Delete the redundant connection nodes in the original dependency path that are broken due to the insertion of the correction point, and re-establish the dependency relationship between the correction point and the adjusted parent node and child nodes to form a cross-level correction dependency path.

5. The digital-based early warning method for industrial chain risk data according to claim 4 is characterized in that: Based on the revised dependency path, the dynamic industry knowledge network is updated and the reconstructed cross-level semantic edges are fused with the original syntactic edges to obtain the comprehensive strength of the fused path, including: Based on the revised dependency paths, dependency types, and strength values, new cross-level semantic edges are added to the dynamic industry semantic network and associated and integrated with the original syntactic edges. If the dependency type of the cross-level semantic edge between the same nodes is consistent with that of the original syntactic edge, they are merged into a single semantic edge, and the strength value is dynamically adjusted according to the dependency strength of the original path and the revised path; If the dependency type of the cross-level semantic edge between the same nodes conflicts with the original syntactic edge, the semantic edge associated with the revised path is retained according to the dependency type rule, and the association strength is recalculated based on the level depth and dependency strength; The integrated semantic edges and association strengths are used as quantitative indicators of the fusion path and are synchronously updated to the dynamic semantic network to obtain the comprehensive strength of the fusion path.

6. The digital-based early warning method for industrial chain risk data according to claim 5 is characterized in that: Based on the comprehensive strength of the fusion path, the risk diffusion simulation is performed through the risk transmission analysis engine to obtain risk quantification results, including: Based on the comprehensive strength and dependency types of fusion paths in the dynamic industry semantic network, a multi-path risk diffusion analysis is performed through the risk transmission analysis engine; Define the risk transmission direction constraint based on the dependency type of the fusion path and convert the comprehensive strength of the path into the transmission probability; Starting from the target detection point, the risk diffusion path along the cross-level semantic edges and original syntactic edges is analyzed. When the risk is transmitted to the branch node, multi-path diversion calculation is performed based on the transmission probability. During the diffusion process, the cumulative risk intensity of each path is dynamically calculated and adjusted based on the attenuation relationship between the path transmission probability and the node level depth; Based on the cumulative intensity and transmission rate of risks reaching the nodes, the real-time risk value of the nodes is quantified to obtain risk quantification results, including the risk value distribution of each node, the identification of key transmission paths and the cross-level diffusion rate.

7. The digital-based early warning method for industrial chain risk data according to claim 6 is characterized in that: Match the risk quantification results with the preset thresholds in the industry risk threshold library, including: According to the industry classification and risk type of the industrial chain nodes, the corresponding risk threshold interval, path transmission rate threshold and key path impact threshold are extracted from the threshold library; If the real-time risk value of a node is greater than the risk threshold of the industry to which it belongs, it will be marked as a high-risk node and an industry chain-level risk warning signal will be generated; Build an early warning report based on industry chain-level risk warning signals, mark the location of high-risk nodes, risk value deviation amplitude and related transmission paths, and identify the diffusion rate and path impact range that are greater than the threshold; Correlate and analyze the risk quantification results with the early warning report, and automatically adjust the threshold intervals and impact determination rules corresponding to the industry classification in the threshold library.

8. A digital-based early warning system for industrial chain risk data, which implements the method according to any one of claims 1 to 7, characterized in that: include: The semantic parsing module is used to extract the semantics of industry chain entities in unstructured text, construct a syntactic tree with the core predicate as the root node, annotate the dependency type and dependency arc strength, and then calculate the semantic weight to identify the final three target detection points of causal influence; The geometric modeling module is used to map the target detection points into a two-dimensional coordinate system, normalize the semantic vector, generate the minimum convex polygon through Delaunay triangulation, calculate the geometric center and project it into the syntax tree structure, and generate the path correction point; The path correction module is used to extract the predicate-argument structure of the path correction point and adjacent nodes, and reconstruct the corrected dependency path; The knowledge fusion module is used to update the dynamic industry knowledge network based on the revised dependency path, fuse the connection strength of cross-level semantic edges with the original syntactic edges, and generate the comprehensive strength of the fused path; The risk simulation module is used to simulate the risk diffusion of the industrial chain based on the comprehensive strength of the fusion path and use the risk transmission analysis engine to obtain risk quantification results; The threshold matching module is used to compare the risk quantification results with the preset thresholds in the industry risk threshold library to automatically determine and output the risk warning level.

9. A computing device, characterized in that include: one or more processors; A storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a program, which, when executed by a processor, implements the method according to any one of claims 1 to 7.

Images