Polymorphic data asset context analysis and industry chain collaboration method

Abstract

The invention discloses a polymorphic data asset context analysis and industry chain collaboration method, and particularly relates to the technical field of data intelligence and industry collaboration, comprising the following steps: accessing multi-source data through a heterogeneous adapter group to generate a four-dimensional intermediate representation containing entities, relationships, timestamps and space coordinates; inputting the representation into a space-time diagram neural network to construct a data association dynamic matrix carrying a dynamic attenuation weight; performing space-time interpolation restoration on the matrix missing link to form a global connected data asset topology network; generating a cross-enterprise security collaboration map based on the differential privacy chain architecture hierarchical encryption topology network; activating the four-dimensional relation chain in response to the collaborative request, and outputting a Nash equilibrium strategy through multi-agent reinforcement learning; an asset value report is generated after the strategy is executed, data distribution drift is detected based on KL divergence, and incremental updating of the data association dynamic matrix is triggered. According to the method, the problems of dynamic modeling and security collaboration of multi-source heterogeneous data are solved, and the decision-making accuracy of an industrial chain is remarkably improved.

Description

Technical Field

[0001] This invention relates to the fields of data intelligence and industrial collaboration technology, and more specifically, to methods for analyzing the context of multi-form data assets and collaborating with the industrial chain. Background Technology

[0002] In the field of supply chain collaboration technology, existing technologies typically use a unified data interface to access multiple heterogeneous data sources. After a standardized cleaning process, this data is imported into a graph database to construct a static entity relationship graph. Based on supply chain path optimization rules or inventory matching rules in a predefined rule engine, the system performs relationship analysis through graph traversal algorithms, generates collaborative strategies such as logistics scheduling plans, and distributes them for execution. Ultimately, the results are evaluated and feedback is assessed through weekly or monthly manual reports, forming an open-loop workflow from data access to manual optimization. This entire process is time-consuming and highly dependent on human intervention.

[0003] However, in practical use, it still has some drawbacks, such as the inability of static graphs to capture spatiotemporal dynamic relationships, leading to model distortion; the use of coarse-grained desensitization in cross-enterprise data collaboration, which neither protects sensitive information hierarchically nor preserves the usability of the relational topology; the rule-driven decision-making mechanism lacks adaptability, requiring a full model reconstruction when data distribution shifts, resulting in excessively long response delays; and due to incomplete data collection or transmission delays, the constructed entity relationship graph contains a large number of missing links, failing to reflect the overall collaborative status of the industry chain, thus affecting data integrity and the reliability of decision analysis. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, this invention provides a method for multi-form data asset context analysis and industry chain collaboration, which addresses the problems raised in the background section through the following solutions.

[0005] To achieve the above objectives, this invention provides the following technical solution: a method for multi-form data asset context analysis and industry chain collaboration, comprising: S1: Access multi-form data assets through a heterogeneous adapter group to generate a four-dimensional intermediate representation; the generated four-dimensional intermediate representation includes entities, relationships, timestamps, and spatial coordinates. S2: Input the four-dimensional intermediate representation into the spatiotemporal graph neural network to construct a data association dynamic matrix carrying dynamic decay weights; S3: Interpolate and repair the missing links in the data association dynamic matrix to output a fully connected data asset topology network; S4: Encrypt and transform the data asset topology network using a differential privacy chain architecture to generate a cross-enterprise secure collaboration graph; S5: Responding to the collaborative decision-making request, activate the four-dimensional relationship chain on the security collaboration graph, and output a cross-enterprise Nash equilibrium strategy through multi-agent reinforcement learning; S6: After executing the Nash equilibrium strategy, generate an asset value report based on the cross-enterprise Nash equilibrium strategy; trigger incremental updates of the data association dynamic matrix based on data distribution drift detection in the report.

[0006] Preferably, the operation of the heterogeneous adapter group includes: Match the corresponding heterogeneous adapter according to the data source type, extract entities and relationships using natural language processing technology, and extract timestamps and spatial coordinates using spatiotemporal data parsing tools; the four-dimensional intermediate representation retains the characteristics of entities as data carriers, relationships as business associations, timestamps recording dynamism, and spatial coordinates reflecting physical locations.

[0007] Preferably, the construction of the data association dynamic matrix includes: The four-dimensional intermediate representation is transformed into a spatiotemporal graph structure, where nodes correspond to entities and edges carry timestamps and spatial coordinates. The dynamic decay weight of each edge is calculated based on the time decay factor and the spatial decay factor. The spatiotemporal graph is input into the spatiotemporal graph neural network, and spatial dependencies are learned through convolutional layers and temporal dependencies are learned through time loop layers. The output is a dynamic data association matrix that quantifies the real-time association strength between entities.

[0008] Preferably, the dynamic decay weight is determined as follows: The time decay factor is calculated based on the difference between the current time and the relation timestamp, as well as the time decay coefficient configured according to industry characteristics; the spatial decay factor is calculated based on the spatial distance between entities and the sensitivity to transportation costs; the final weight is generated by combining the time decay factor and the spatial decay factor.

[0009] Preferably, the interpolation repair includes: A threshold for determining missing links is set, and the repair value is calculated based on the average weight of adjacent timestamps and the average product of the weights of indirect links within the spatial neighborhood, combined with the spatiotemporal weight coefficient. The distance threshold of the spatial neighborhood is set according to the industry average cooperation radius, and the product of the path weights of the intermediate entities for indirect links reflects the association strength.

[0010] Preferably, the spatiotemporal weighting coefficient prioritizes strengthening the contribution of the time dimension while retaining the supplementary role of spatial neighborhood information, ensuring that the repair value conforms to the business characteristics of spatiotemporal correlation; the fully connected data asset topology network requires that there be a non-zero correlation path between any two entities.

[0011] Preferably, the hierarchical encryption of the differential privacy chain architecture includes: Entity attributes are divided into three levels: high sensitivity, medium sensitivity, and low sensitivity. High-intensity Laplace noise is added to high-sensitivity data, Gaussian noise is added to medium-sensitivity data, and slight noise is added to low-sensitivity data. The relationships between encrypted entities are preserved through the consistency of the added noise, generating a cross-enterprise secure collaboration graph.

[0012] Preferably, activating the four-dimensional relationship chain includes: Based on the decision-making objective, extract the target entity set, core relationship types, decision time window, and spatial association range; map each entity as an independent intelligent agent, whose state vector contains desensitized attributes, association strength, and decision indicators; balance individual gains with the overall benefits of the industry chain through a reward function, and train the policy set until it satisfies the Nash equilibrium condition using a deep deterministic policy gradient algorithm.

[0013] Preferably, the reward function is designed as a weighted sum of individual income and overall industry chain benefits, with the weights adjusted by a synergy coefficient; the Nash equilibrium strategy ensures that each entity's strategy is the optimal response to the strategies of other entities.

[0014] Preferably, the incremental update includes: The asset value of entity data is calculated based on the asset value report, which is jointly determined by the basic value, the association weight, and the synergistic value-added. Data drift is detected by the distribution difference metric. If the value change exceeds the threshold, the entity and its associated entities that exceed the limit are located. The dynamic decay weight is recalculated only for the affected entity pairs, and the corresponding items in the data association dynamic matrix are updated.

[0015] The technical effects and advantages of this invention are as follows: 1. Accurately capture spatiotemporal dynamic relationships and improve model realism: By generating a four-dimensional intermediate representation containing entities, relationships, timestamps and spatial coordinates, and combining it with a spatiotemporal graph neural network to construct a dynamic data association matrix with dynamically decaying weights, the problem that traditional static graphs cannot capture spatiotemporal dynamic relationships of data is effectively solved. 2. Achieve hierarchical security collaboration while balancing privacy protection and connectivity availability: The differential privacy chain architecture is used to perform hierarchical encryption on the data asset topology network. Different levels of noise are added to high, medium and low sensitive data respectively. While preventing the leakage of sensitive information, the consistency of noise addition preserves the relationship between entities. 3. Adaptive decision-making mechanism to balance individual and overall interests: Based on multi-agent reinforcement learning, cross-enterprise Nash equilibrium strategies are solved on the security collaboration graph. Each entity is mapped as an independent agent. By designing a reward function that takes into account both individual benefits and the overall benefits of the industrial chain, the output strategy satisfies the Nash equilibrium condition. 4. Efficient incremental updates reduce response latency: Data distribution drift is detected by KL divergence. When a significant drift is detected, only the weights of the affected entities and related links are incrementally updated, rather than the entire model is reconstructed. 5. Improve data integrity and build a fully connected topology: Repair missing links in the dynamic matrix of data association through spatiotemporal interpolation, estimate missing values ​​based on the average weight of adjacent timestamps and the average product of the weights of indirect links in spatial neighborhoods, and form a fully connected data asset topology network without isolated nodes. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall structure of the present invention.

[0017] Figure 2 This is a schematic diagram illustrating the construction of the dynamic data association matrix according to the present invention.

[0018] Figure 3 This is a schematic diagram of the differential privacy hierarchical encryption mechanism of the present invention. Detailed Implementation

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

[0020] refer to Figures 1-3 The multi-form data asset context analysis and industry chain collaboration methods shown include: S1: Access multi-form data assets through a heterogeneous adapter group to generate a four-dimensional intermediate representation; the generated four-dimensional intermediate representation includes entities, relationships, timestamps, and spatial coordinates. S2: Input the four-dimensional intermediate representation into the spatiotemporal graph neural network to construct a data association dynamic matrix carrying dynamic decay weights; S3: Interpolate and repair the missing links in the data association dynamic matrix to output a fully connected data asset topology network; S4: Encrypt and transform the data asset topology network using a differential privacy chain architecture to generate a cross-enterprise secure collaboration graph; S5: Responding to the collaborative decision-making request, activate the four-dimensional relationship chain on the security collaboration graph, and output a cross-enterprise Nash equilibrium strategy through multi-agent reinforcement learning; S6: After executing the Nash equilibrium strategy, generate an asset value report based on the cross-enterprise Nash equilibrium strategy; trigger incremental updates of the data association dynamic matrix based on data distribution drift detection in the report.

[0021] S1: Access multi-form data assets through a heterogeneous adapter group to generate a four-dimensional intermediate representation; the generated four-dimensional intermediate representation includes entities, relationships, timestamps, and spatial coordinates. The information extracted by the adapter is standardized to generate a unified format four-dimensional intermediate representation, formally represented as follows: ; Select Once converted to a data entity, its formal representation is as follows: , For the relationship between entities, For timestamps, Spatial coordinates; In the process of heterogeneous data federation access, the first step is to select suitable heterogeneous adapters based on the data source type. For example, database adapters are used for databases, and text adapters are used for text data. These adapters are then used to access multi-form data assets. Next, after the data is accessed, natural language processing technology (named entity recognition to extract entities from text, relation extraction models to identify relationships between entities) and spatiotemporal data parsing tools (time stamps are extracted based on regular expressions or time parsing libraries, and addresses are converted into spatial coordinates through geocoding services) are used to extract key information such as entities, relationships, timestamps, and spatial coordinates from the data. These are then integrated to construct a four-dimensional intermediate representation, completing the conversion from multi-source heterogeneous data to a unified intermediate format.

[0022] It should be further explained that the four-dimensional intermediate representation is a spatiotemporal fusion data structure customized for federated heterogeneous data. It forms a unified machine semantic layer that supports dynamic decay modeling by forcibly mapping four basic dimensions: entities, relationships, timestamps, and spatial coordinates. It is the core technical hub connecting data access and contextual analysis. The design significance of the four-dimensional intermediate representation is to retain the full-dimensional characteristics of the data: entities are the core carriers of data, relationships reflect the business connections between entities, timestamps record the dynamics of relationships, and spatial coordinates reflect physical connections.

[0023] S2: Input the four-dimensional intermediate representation into the spatiotemporal graph neural network to construct a data association dynamic matrix carrying dynamic decay weights; Transform the four-dimensional intermediate representation into a spacetime diagram: ,in: Node set Corresponds to all entities; Edge set For the relationships between entities, each edge carries a timestamp. (Time of first establishment or most recent activity of the relationship) and spatial coordinates (The geographical location where the relationship occurred); Dynamic decay weights are used to define edges The decay weight needs to take into account both time decay and spatial decay: Time decay factor: ,in for The time difference from now, For the current time, Let be the time decay coefficient, and ;like (Future time), default (Relationships that have not yet occurred are not considered at this time); Spatial attenuation factor: ,in For entities and Spatial distance ( (The spatial coordinates of the two entities) Let be the spatial attenuation coefficient, and ; By combining the attenuated intensity over time and space, a dynamic attenuation weight is obtained for each edge: ; The spatiotemporal graph is input into the spatiotemporal graph neural network. Convolutional layers and recurrent layers learn the dynamic dependencies between entities and construct a dynamic matrix of data associations. : , representing entities arrive The dynamic correlation strength (including attenuation).

[0024] Among them, the time decay coefficient This parameter, which can be set according to industry characteristics, quantifies the natural weakening of the relationship strength between entities over time. It is calculated together with the difference between the current time and the relationship timestamp to determine the time decay factor, ultimately affecting the dynamic weight of entity associations; spatial decay coefficient. Similarly, it is a parameter configured based on the industry's sensitivity to transportation costs, used to measure the impact of spatial distance between entities on the strength of association. It is used together with the actual spatial distance between entities to calculate the spatial attenuation factor and participate in the comprehensive calculation of dynamic attenuation weight. It should be further explained that in the spatiotemporal graph neural network, the convolutional layer is used to learn the spatial dimension dependency, and the temporal recurrent layer is used to learn the temporal dimension dependency. The combination of the two can capture the dynamic relationship between a certain enterprise and other enterprises at a specific time and in a specific region. The final generated data association dynamic matrix can quantify the real-time association strength between entities.

[0025] S3: Interpolate and repair missing links in the data association dynamic matrix to output a fully connected data asset topology network; Due to incomplete data collection or transmission delays, there may be missing values ​​in the data association dynamic matrix. However, since a relationship actually exists, time-space interpolation is used to repair missing links and ensure full connectivity of the topology network. The steps are as follows: Set threshold (e.g., 0.05), when When this occurs, it is identified as a potentially missing link; For missing links Weights are estimated based on their spatiotemporal neighborhood information: Take adjacent timestamps Midlink Weighted mean: ; Take entities within the spatial neighborhood (satisfy and The mean of the indirect link weight product: ; Final repair value ,in For the spatiotemporal weighting coefficients, and .

[0026] The repaired matrix forms a data asset topology network without isolated nodes. There exists a non-zero association path between any two entities.

[0027] in, The threshold is set by combining data characteristics, historical experience and business needs. If the threshold is greater than 0.2, it may misjudge the actual weak correlation as a missing link, resulting in over-repair; if the threshold is less than 0.01, it may miss the real missing link. It can filter out links that should theoretically be related but are not reflected due to data issues. After repair, it can more realistically reflect the actual collaborative status of the industry chain, providing a reliable foundation for subsequent construction and decision-making of security collaboration maps.

[0028] Spatiotemporal weighting coefficient This is because, in the correlation analysis of supply chain data assets, the time dimension usually has a more significant impact on the weight of the links; the timeliness of the relationships between entities is crucial to the dynamism of the data correlation dynamic matrix, while the spatial dimension must also be taken into account to ensure the comprehensiveness of the repair. The value of balancing the contributions of time and space factors highlights the core role of time decay while not ignoring the supplementary value of spatial neighborhood information, making the repair value of missing links more in line with the real characteristics of spatiotemporal correlation in actual business.

[0029] It should be further explained that adjacent timestamps usually refer to the time windows before and after the missing link time point to ensure the continuity of the time dimension; the distance threshold of the spatial neighborhood can be set according to the industry average cooperation radius; the indirect link refers to the path from target entity A to entity E to target entity B, and its weight product (the weight from A to E multiplied by the weight from E to B) can reflect the strength of the indirect association between A and B.

[0030] S4: The data asset topology network is encrypted and transformed using a differential privacy chain architecture to generate a cross-enterprise security collaboration graph; The entity attributes in the topology are divided into three levels: high sensitivity, medium sensitivity, and low sensitivity. Apply differential privacy chain architecture to different levels of data: Highly sensitive data: Add Laplace noise The formula is: Privacy budget (High noise level); Medium-sensitive data: Add Gaussian noise The formula is: , It follows a Gaussian distribution. The variance is Gaussian noise. Low-sensitivity data: Add slight noise The formula is: ,in, The variance is slightly noisy, and It only obscures data details without affecting the overall recognition.

[0031] After the above encryption process, the entity attribute values ​​in the data asset topology are obfuscated, but the relationships between entities are preserved through the consistency of noise addition (i.e., if the entities in the original topology are not obfuscated). With entity The association exists, and it is preserved after encryption; only the weight values ​​are proportionally increased with noise. This results in a cross-enterprise security collaboration graph. This not only meets the needs of data sharing between enterprises, but also prevents the leakage of sensitive information through differential privacy technology.

[0032] Highly sensitive data includes core data such as monetary amounts and customer information; moderately sensitive data includes operational data such as supply and transportation; and lowly sensitive data includes publicly available data such as names and addresses.

[0033] It should be further explained that Laplace noise is the best at protecting highly sensitive numerical data in the differential privacy chain architecture, as its noise distribution is concentrated near the true value and can effectively mask the specific value; Gaussian noise is suitable for moderately sensitive data, with a more gradual noise distribution, which can protect privacy while preserving the data trend; slight noise in low-sensitivity data only masks details and does not affect the overall recognition.

[0034] S5: Responding to the collaborative decision-making request, activate the four-dimensional relationship chain on the security collaborative graph, and output a cross-enterprise Nash equilibrium strategy through multi-agent reinforcement learning; Based on the decision-making objectives, from The four-dimensional relationship chain of related entities is extracted and formally represented as: ;in For the set of entities related to the target, This is the core relationship type between entities. For decision-making time windows (e.g.) ,in The decision-making start time, (Decision deadline) This refers to the spatial association range.

[0035] After activating the four-dimensional relationship chain, the cross-enterprise Nash equilibrium strategy is solved through multi-agent reinforcement learning. The specific steps are as follows: Map each entity in the four-dimensional relation chain to an independent intelligent agent, i.e., a set of intelligent agents. The number of intelligent agents Each intelligent agent corresponds to a decision-making body of an enterprise; Each agent The state vector includes: self-de-sensitization attributes: The strength of association with other agents (derived from the edge weights of the secure collaboration graph): and Indicators related to decision-making objectives: ; The state vector is formalized as follows: .

[0036] The actions of intelligent agents are collaborative operations aimed at decision-making objectives, such as adjusting supply quotas, opening specific data interfaces, and changing cooperation prices.

[0037] The reward function needs to balance individual gains with the overall benefits of the industry chain (the gains of each agent): ,in For intelligent agents Individual benefits, For the overall benefit of the industrial chain, Let be the synergy coefficient, and It is used to adjust the weight of individual and overall interests.

[0038] The agent is trained using the Deep Deterministic Policy Gradient (DDPG) algorithm, and the final output is a policy set. It satisfies the Nash equilibrium condition: This means that each firm's strategy is the optimal response to the strategies of other firms.

[0039] Among them, the activation of the four-dimensional relationship chain provides a precise decision boundary for agent training. The target-related entity set, core relationship type, time window and spatial association range included in the chain directly limit the decision object, association dimension and scope of action of the agent.

[0040] It needs to be further explained that the process of the policy set converging to Nash equilibrium during the training of the deep deterministic policy gradient algorithm relies on the precise quantification of the balance between individual and overall interests by the reward function. The final output policy set satisfies the Nash equilibrium condition. Its essence is that each agent forms a mutually adapted optimal response strategy after comprehensively considering its own state, the strength of its association with other agents, and the decision target indicators, so as to ensure the balance between individual gains and the overall benefits of the industry chain.

[0041] S6: After executing the Nash equilibrium strategy, an asset value report is generated based on the cross-enterprise Nash equilibrium strategy; based on the data distribution drift detection in the report, an incremental update of the data association dynamic matrix is ​​triggered.

[0042] After implementing the Nash equilibrium strategy, the value of data assets of each entity in the industry chain is quantitatively assessed, generating an asset value report. The core indicator of the report is the value of the entity's data assets. The calculation formula is: ,in For entities The fundamental value of one's own data (such as data volume and scarcity score). For entities in the data association dynamic matrix With entity Association weights For entities The added value generated through collaboration with other entities.

[0043] Based on the data distribution in the asset valuation report, we detect whether significant drift has occurred. We use KL divergence to measure the difference in data distribution before and after strategy implementation. ,in For the data distribution after the strategy is executed (such as the latest data distribution) or distributed), The baseline data distribution before strategy execution; if ,in If the preset threshold is used, it is determined that the data distribution has shifted significantly.

[0044] When data distribution drift is detected, an incremental update of the data association dynamic matrix is ​​triggered, adjusting only the weights of the links affected by the drift, rather than a full reconstruction: Location drift associated entities: Filter out entity and its related entities ,in The threshold for value change; Recalculate the dynamic decay weights: for the above entity pairs Based on the latest timestamp and spatial coordinates Update time decay factor and spatial decay factor ( (The latest spatial distance); Update the dynamic matrix of data association: the affected weights are updated as follows The unaffected weights remain unchanged.

[0045] in, This is based on the need to balance the accuracy of data distribution drift detection with the dynamic adaptability of the industry chain. KL divergence is used to measure the difference in data distribution before and after strategy execution. A threshold of 0.1 can effectively distinguish between normal fluctuations and significant drifts: if the difference is less than 0.1, it means that the change in data distribution is within an acceptable range and there is no need to update the data association dynamic matrix; if it exceeds 0.1, it is judged as a significant drift and an incremental update needs to be triggered. This value avoids frequent updates caused by small fluctuations and can capture substantial changes that are sufficient to affect the collaborative decision-making of the industry chain in a timely manner, ensuring that the data association dynamic matrix always keeps in line with the actual state of the industry chain and provides a reliable basis for subsequent collaborative decision-making. In daily operations, the value change caused by short-term fluctuations is usually within 0.1 to 0.15, while changes exceeding 0.2 are more likely to reflect substantial business adjustments, which is consistent with significantly drifting business perceptions. threshold and By forming a progressive relationship on a scale of magnitude, noise can be filtered collaboratively, ensuring that weights are updated only for entities that truly affect the industrial chain, thus meeting the efficiency requirements of incremental updates in closed-loop feedback.

[0046] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for analyzing the context of multi-form data assets and promoting supply chain collaboration, characterized in that... include: S1: Access multi-form data assets through a heterogeneous adapter group to generate a four-dimensional intermediate representation; the generated four-dimensional intermediate representation includes entities, relationships, timestamps, and spatial coordinates. S2: Input the four-dimensional intermediate representation into the spatiotemporal graph neural network to construct a data association dynamic matrix carrying dynamic decay weights; S3: Interpolate and repair the missing links in the data association dynamic matrix to output a fully connected data asset topology network; S4: Encrypt and transform the data asset topology network using a differential privacy chain architecture to generate a cross-enterprise secure collaboration graph; S5: Responding to the collaborative decision-making request, activate the four-dimensional relationship chain on the security collaboration graph, and output a cross-enterprise Nash equilibrium strategy through multi-agent reinforcement learning; S6: After executing the Nash equilibrium strategy, generate an asset value report based on the cross-enterprise Nash equilibrium strategy; trigger incremental updates of the data association dynamic matrix based on data distribution drift detection in the report.

2. The method for multi-form data asset context analysis and industry chain collaboration according to claim 1, characterized in that, The operation of the heterogeneous adapter group includes: Match the corresponding heterogeneous adapter according to the data source type, extract entities and relationships using natural language processing technology, and extract timestamps and spatial coordinates using spatiotemporal data parsing tools; the four-dimensional intermediate representation retains the characteristics of entities as data carriers, relationships as business associations, timestamps recording dynamism, and spatial coordinates reflecting physical locations.

3. The method for multi-form data asset context analysis and industry chain collaboration according to claim 1, characterized in that, The construction of the dynamic data association matrix includes: The four-dimensional intermediate representation is transformed into a spatiotemporal graph structure, where nodes correspond to entities and edges carry timestamps and spatial coordinates. The dynamic decay weight of each edge is calculated based on the time decay factor and the spatial decay factor. The spatiotemporal graph is input into the spatiotemporal graph neural network, and spatial dependencies are learned through convolutional layers and temporal dependencies are learned through time loop layers. The output is a dynamic data association matrix that quantifies the real-time association strength between entities.

4. The method for multi-form data asset context analysis and industry chain collaboration according to claim 3, characterized in that, The dynamic decay weight is determined as follows: The time decay factor is calculated based on the difference between the current time and the relation timestamp, as well as the time decay coefficient configured according to industry characteristics; the spatial decay factor is calculated based on the spatial distance between entities and the sensitivity to transportation costs; the final weight is generated by combining the time decay factor and the spatial decay factor.

5. The method for multi-form data asset context analysis and industry chain collaboration according to claim 1, characterized in that, The interpolation repair includes: A threshold for determining missing links is set, and the repair value is calculated based on the average weight of adjacent timestamps and the average product of the weights of indirect links within the spatial neighborhood, combined with the spatiotemporal weight coefficient. The distance threshold of the spatial neighborhood is set according to the industry average cooperation radius, and the product of the path weights of the intermediate entities for indirect links reflects the association strength.

6. The method for multi-form data asset context analysis and industry chain collaboration according to claim 5, characterized in that, The spatiotemporal weighting coefficient prioritizes the contribution of the time dimension while retaining the supplementary role of spatial neighborhood information, ensuring that the repair value conforms to the business characteristics of spatiotemporal correlation; the fully connected data asset topology network requires that there be a non-zero correlation path between any two entities.

7. The method for multi-form data asset context analysis and industry chain collaboration according to claim 1, characterized in that, The hierarchical encryption of the differential privacy chain architecture includes: Entity attributes are divided into three levels: high sensitivity, medium sensitivity, and low sensitivity. High-intensity Laplace noise is added to high-sensitivity data, Gaussian noise is added to medium-sensitivity data, and slight noise is added to low-sensitivity data. The relationships between encrypted entities are preserved through the consistency of the added noise, generating a cross-enterprise secure collaboration graph.

8. The method for multi-form data asset context analysis and industry chain collaboration according to claim 1, characterized in that, The activation of the four-dimensional relationship chain includes: Based on the decision-making objective, extract the target entity set, core relationship types, decision time window, and spatial association range; map each entity as an independent intelligent agent, whose state vector contains desensitized attributes, association strength, and decision indicators; balance individual gains with the overall benefits of the industry chain through a reward function, and train the policy set until it satisfies the Nash equilibrium condition using a deep deterministic policy gradient algorithm.

9. The method for multi-form data asset context analysis and industry chain collaboration according to claim 8, characterized in that, The reward function is designed as a weighted sum of individual income and overall industry chain benefits, with the weights adjusted by a synergy coefficient; the Nash equilibrium strategy ensures that each entity's strategy is the optimal response to the strategies of other entities.

10. The method according to claim 1, characterized in that, The incremental update includes: The asset value of entity data is calculated based on the asset value report, which is jointly determined by the basic value, the association weight, and the synergistic value-added. Data drift is detected by the distribution difference metric. If the value change exceeds the threshold, the entity and its associated entities that exceed the limit are located. The dynamic decay weight is recalculated only for the affected entity pairs, and the corresponding items in the data association dynamic matrix are updated.

Images