An artificial intelligence-based digital asset management method and system

Abstract

The application provides a digital asset management method and system based on artificial intelligence, comprising: constructing a dynamic asset image by extracting feature vectors from multiple dimensions of digital assets; constructing an asset semantic correlation network with the dynamic asset image as a node; inputting a knowledge graph into a risk assessment model, combining the dynamic change trend of the dynamic asset image with the abnormal propagation path in the asset semantic correlation network, quantitatively scoring the compliance risk, value decay risk and security threat of each digital asset, and generating an asset risk situation atlas; according to the asset risk situation atlas, generating an asset management strategy for assets of different risk levels by a strategy generation model, and executing the strategy through a unified management interface; collecting asset state change data after the execution of the asset management strategy as a feedback signal, and incrementally learning and updating the feature extraction model and the risk assessment model. The application enhances the intelligent level and security protection capability of digital asset management.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and specifically to a digital asset management method and system based on artificial intelligence. Background Technology

[0002] In recent years, with the deepening of digital transformation and the rapid popularization of artificial intelligence technology, the scale of digital assets generated in various industries has grown exponentially, and digital assets have become an important component of core production factors for enterprises. Existing digital asset management methods are mainly based on manual metadata annotation, rule matching, and relational database retrieval. These methods can meet basic needs in the early stages when digital assets are small in scale and of simple type, but existing solutions have problems such as isolated asset management and lagging risk assessment. Summary of the Invention

[0003] Based on the above-mentioned problems, this invention proposes a digital asset management method and system based on artificial intelligence. Through this invention, the accuracy of risk identification and the adaptability of strategies are continuously improved, and the intelligence level and security protection capability of digital asset management are significantly enhanced.

[0004] In view of this, one aspect of the present invention proposes a digital asset management method based on artificial intelligence, comprising: By using a multimodal feature extraction model, feature vectors are extracted from the dimensions of content semantics, behavioral trajectory, association relationship and life cycle status of digital assets, and a dynamic asset profile that can be updated in real time as the asset status changes. Using the dynamic asset profile as nodes, an asset semantic association network is constructed based on the semantic similarity, ownership relationship and business dependency relationship between assets, and the asset semantic association network is stored in a knowledge graph; The knowledge graph is input into the risk assessment model, and combined with the dynamic change trend of the dynamic asset profile and the abnormal propagation path in the asset semantic association network, the compliance risk, value decay risk and security threat of each digital asset are quantitatively scored to generate an asset risk situation map. Based on the asset risk profile, the strategy generation model generates asset management strategies for assets with different risk levels, and then issues and executes them through a unified management interface. Data on changes in asset status after the implementation of asset management strategies are collected as feedback signals to incrementally learn and update the feature extraction model and risk assessment model.

[0005] Optionally, the step of using a multimodal feature extraction model to extract feature vectors from the dimensions of content semantics, behavioral trajectory, association relationship, and lifecycle state of digital assets, and constructing a dynamic asset profile that can be updated in real time as the asset state changes, includes: The collected digital assets are automatically identified and classified according to their data modalities, and text, image, structured data and behavior log assets are respectively diverted to the corresponding single-modal feature extraction sub-models in the multimodal feature extraction model; Each single-modal feature extraction sub-model extracts content semantic vectors, behavioral trajectory sequence vectors, relational topology vectors, and lifecycle state vectors from the corresponding assets to form local feature vectors for each modality. The local feature vectors of each modality are input into the cross-modal attention fusion module, and adaptive weighted fusion is performed based on the correlation weights between features of different modalities to output a unified multimodal comprehensive feature vector; Using multimodal integrated feature vectors as the core and combining asset metadata, an initial dynamic profile of the digital asset is generated, and the profile is stored in the profile library in an incrementally updatable structure. When any modal feature of an asset is detected to change, the corresponding single-modal sub-model is triggered to re-extract local features, and the corresponding portrait in the portrait library is locally refreshed through the cross-modal attention fusion module.

[0006] Optionally, the step of constructing an asset semantic association network based on the dynamic asset profile as nodes, according to the semantic similarity, ownership relationship, and business dependency relationship between assets, and storing the asset semantic association network in a knowledge graph includes: Three types of relationships between assets are predefined, including semantic similarity relationships, ownership relationships, and business dependency relationships. Based on these three types of relationships and their constraint rules, a digital asset domain ontology is constructed as a unified semantic framework for subsequent generation of relationship networks and storage of knowledge graphs. Using the multimodal comprehensive feature vector of each dynamic asset profile as input, the semantic similarity calculation model is used to measure the similarity between each profile. When the similarity exceeds the adaptive threshold, semantic similarity edges are generated between the corresponding profile nodes. Based on three types of relationships, the ownership relationship and business dependency relationship between assets are automatically extracted from asset metadata and business system configuration information, and converted into directed relationship edges of the corresponding types to obtain ownership relationship edges and business dependency relationship edges. By integrating semantic similarity edges with attribution and business dependency edges, and using dynamic asset profiles as common nodes, an asset semantic association network containing multiple types of heterogeneous edges is constructed. Based on the domain ontology, the nodes and various relation edges in the asset semantic association network are mapped to entity and attribute relation triples in the knowledge graph and persistently stored in the graph database. When any dynamic asset portrait in the portrait library is triggered to be updated, the entity node corresponding to this dynamic asset portrait in the knowledge graph is taken as the update starting point. The semantic similarity calculation, relation edge extraction, heterogeneous network fusion and triple mapping storage operations involving the entity node are re-executed to incrementally update the entity node and its related relation triples in the knowledge graph.

[0007] Optionally, the step of inputting the knowledge graph into the risk assessment model, combining the dynamic changing trends of the dynamic asset profile with the abnormal propagation paths in the asset semantic association network, to quantitatively score the compliance risks, value decay risks, and security threats of each digital asset, and generate an asset risk situation map, includes: Time-series modeling is performed on the historical update sequence of each dynamic asset portrait in the portrait library to extract the portrait change trend features of each dynamic asset portrait in terms of content semantics, behavior trajectory, relationship and life cycle status. Using a knowledge graph as input and the entity nodes corresponding to each digital asset in the knowledge graph as the operation objects, the features of the neighboring entity nodes of each entity node are aggregated through a graph neural network. The aggregated node representation is compared with the baseline representation under the normal association mode to identify abnormal entity nodes that deviate from the normal association mode. The abnormal propagation path of abnormal entity nodes spreading to neighboring entity nodes along the affiliation relationship edge and the business dependency relationship edge is tracked to locate the potential links of risk spread. By inputting the characteristics of profile change trends and abnormal propagation paths into the risk assessment model, each digital asset is independently quantitatively scored from three dimensions: compliance risk, value decay risk, and security threat, resulting in a three-dimensional risk score vector for each asset. Based on the characteristics of abnormal propagation paths, the three-dimensional risk score vectors of related assets in the propagation chain are transmitted and corrected so that the risk scores reflect the superposition or amplification effect of mutual influence between assets. Using a knowledge graph as the topological framework, the corrected three-dimensional risk score vectors of each asset are used as node attributes, and the anomaly propagation path is used as the risk transmission edge attribute. These are overlaid and rendered onto the knowledge graph to generate an asset risk situation map that intuitively presents the risk level of each asset and the risk transmission relationship. Continuously monitor changes in the profile database and knowledge graph. When any dynamic asset profile or knowledge graph relationship is updated, the asset risk situation graph is incrementally refreshed to achieve continuous and dynamic perception of the risk situation.

[0008] Optionally, the step of generating asset management strategies for assets of different risk levels based on the asset risk profile using a strategy generation model, and then issuing and executing these strategies through a unified management interface, includes: Based on the three-dimensional risk scoring vector of each asset in the asset risk situation map, and according to the preset risk level classification rules, all digital assets are classified into four levels: high risk, medium risk, low risk and normal, and the archiving results are written into the asset risk level ledger. Using the risk level and three-dimensional risk score vector of each asset as input, a set of candidate management strategies that match the current risk characteristics is retrieved from a pre-built strategy candidate library. The candidate strategies cover types such as access control tightening, encryption protection upgrade, backup migration and deregistration. The candidate management strategy set, together with the risk transmission edge attributes of the corresponding assets in the asset risk situation map, are input into the strategy generation model. The strategy generation model comprehensively evaluates the suppression effect of each candidate strategy on the risk of the current asset and its related assets, and outputs the optimal management strategy for each asset. Perform global conflict detection on the optimal management strategy for all assets, identify strategy execution conflicts caused by business dependencies between assets, adjust conflicting strategies through priority coordination rules, and output a final set of conflict-free strategy executions. The final policy execution set is arranged according to asset ownership domain and policy type, and execution instructions are issued to the corresponding asset management system through a unified management interface to trigger the automated execution of access control policy updates, encryption parameter configuration, backup migration tasks or deregistration processes. Collect the results and status of each strategy after execution, write the results back to the asset risk level ledger, update the life cycle management record of the corresponding asset, and form a closed-loop management of the entire life cycle from risk perception, strategy decision-making to execution and write-off.

[0009] Optionally, the step of using data on changes in asset status after the execution of the asset management strategy as feedback signals to incrementally learn and update the feature extraction model and the risk assessment model includes: Using the dynamic asset profile and asset risk profile map before the strategy is issued and executed as a baseline snapshot, the profile features and risk scores of the corresponding assets are re-collected after the strategy is executed, and the state difference before and after execution is calculated to obtain a structured feedback signal that reflects the actual effect of the strategy. The feedback signals are filtered to remove abnormal state change samples caused by external interference, and retain valid feedback samples that can truly reflect the causal effect of the strategy. Based on the execution result write-back records in the strategy execution result ledger, the valid samples are automatically labeled with positive and negative effect tags. Using the changes in asset profile features in the effective feedback samples as incremental training data, the multimodal feature extraction sub-model and the cross-modal attention fusion module are incrementally learned, so that the feature extraction model can continuously adapt to the evolution of asset content, behavior and relationships. Using the risk score changes and abnormal propagation path correction information in the effective feedback samples as incremental training data, the graph neural network and risk assessment model are incrementally learned to correct the model's assessment bias of compliance risk, value decay risk and security threat. Based on positive and negative effect labels, the applicable conditions and priority weights of corresponding strategies in the strategy candidate library are adjusted, and the strategy generation model is incrementally fine-tuned with effective feedback samples, so that the strategy decision-making ability can be continuously improved as the asset environment evolves. After each round of incremental learning, the updated models are evaluated using the retained validation sample set. When the evaluation metric fails to reach the preset improvement threshold, a retraining process with an expanded sample range is triggered to ensure the effectiveness of incremental learning and prevent performance degradation of the model due to low-quality feedback.

[0010] Optionally, when measuring the pairwise similarity of profiles using a semantic similarity calculation model, the cross-domain semantic similarity between any two digital assets a and b is... Calculated by the following formula:

[0011] in, These are the multimodal integrated feature vectors of assets a and b, respectively. The cosine similarity function measures the degree of similarity between two assets in the semantic content dimension. These are the category code vectors for assets a and b in the business classification system, respectively. For Euclidean distance, This is the smoothing coefficient for the class distance; Let be the shortest number of hops between assets a and b in the knowledge graph. The threshold for the number of hops. For indicator functions; These are the sets of operation types that assets a and b have appeared in the historical behavior trajectory sequence, respectively. Jaccard similarity for the historical operation types of two assets is used to measure the degree of commonality between the two assets in the behavioral pattern dimension. These are the coded values ​​representing the current lifecycle stages of assets a and b, respectively. For life cycle stage gap, As a smoothing coefficient for the life cycle stage gap, this exponential decay term gives assets with similar life cycle stages a higher behavioral similarity bonus, avoiding the incorrect classification of assets in different life cycle stages as strongly correlated due to the overlap of historical behavior sets. The learnable balance coefficients for the four similarity components satisfy... , .

[0012] Optionally, when performing three-dimensional risk quantification scoring on each digital asset, the security threat score of asset a is... Calculated by the following formula:

[0013] in, It is a set of neighboring assets in the knowledge graph that have business dependencies or ownership relationships with asset a; The probability of a basic security threat to the neighboring asset n itself; The edge weights for transmitting security threats from asset n to asset a are determined by the degree of anomaly of the corresponding knowledge graph relation edges; This is the propagation path protection coefficient between asset n and asset a, with a value range of (0,1]. It reflects the strength of the existing access control and encryption protection measures between the two assets in suppressing threat propagation. The more comprehensive the protection measures, the stronger the protection. The smaller the value, the lower the actual threat contribution of this transmission path; The vulnerability coefficient of asset a itself is determined by the aging degree of the lifecycle state vector in the asset profile and the completeness of the security configuration. A function representing the magnitude of change in the profile of asset a. This represents the change in the current image feature vector relative to the previous time step. for A subset of neighboring assets whose current security threat score exceeds the high-risk threshold. Its quantity; Let be the sensitivity coefficient of asset a to the number of high-risk neighboring assets. The logarithmic term is used to characterize the nonlinear amplification effect of the increase in the number of high-risk neighboring assets on the security threat of asset a. The logarithmic form is used to prevent the score from expanding infinitely when there are too many high-risk neighboring assets.

[0014] Optionally, when the strategy generation model makes optimization decisions on each candidate management strategy, asset a is in the candidate strategy set. Optimal Strategy The choice is determined by the following formula:

[0015] in, Let t be the expected suppression return of strategy t on the three-dimensional risk score vector of asset a itself; The expected cascading suppression of returns on neighboring related assets after the execution of strategy t via risk transmission edge; The resource cost for executing strategy t is calculated by comprehensively considering storage, computing, and business interruption costs. The expected conflict penalty for strategy t that conflicts with other decided strategies in the current asset risk level ledger; The historical success rate weighted return of strategy t is obtained by the weighted product of the historical success rate of strategy t on assets of the same risk level and the corresponding historical suppression return in the strategy candidate library. It is used to guide strategy decision-making to prioritize well-validated high success rate strategies. Let strategy t be the expected duration of interruption caused by normal business access to asset a. This is the service interruption tolerance threshold. The smoothing coefficient for the interruption duration penalty is used when the interruption duration exceeds a threshold. The strategy of imposing a non-linear, incremental penalty on the score to strongly constrain excessive impact on business continuity was selected. The indicator function is used to activate this penalty term; These are the learnable balancing weight coefficients for each item.

[0016] Another aspect of the present invention provides an artificial intelligence-based digital asset management system for executing an artificial intelligence-based digital asset management method, comprising: a server and an asset management terminal; The server is configured as follows: By using a multimodal feature extraction model, feature vectors are extracted from the dimensions of content semantics, behavioral trajectory, association relationship and life cycle status of digital assets, and a dynamic asset profile that can be updated in real time as the asset status changes. Using the dynamic asset profile as nodes, an asset semantic association network is constructed based on the semantic similarity, ownership relationship and business dependency relationship between assets, and the asset semantic association network is stored in a knowledge graph; The knowledge graph is input into the risk assessment model, and combined with the dynamic change trend of the dynamic asset profile and the abnormal propagation path in the asset semantic association network, the compliance risk, value decay risk and security threat of each digital asset are quantitatively scored to generate an asset risk situation map. Based on the asset risk profile, the strategy generation model generates asset management strategies for assets with different risk levels, and then issues and executes them through a unified management interface. Data on changes in asset status after the implementation of asset management strategies are collected as feedback signals to incrementally learn and update the feature extraction model and risk assessment model.

[0017] The technical solution of this invention provides an artificial intelligence-based digital asset management method comprising: utilizing a multimodal feature extraction model to extract feature vectors from the dimensions of content semantics, behavioral trajectory, association relationships, and lifecycle status of digital assets, constructing a dynamic asset profile that can be updated in real time as the asset status changes; using the dynamic asset profile as nodes, constructing an asset semantic association network based on the semantic similarity, ownership relationship, and business dependency relationship between assets, and storing the asset semantic association network in a knowledge graph; inputting the knowledge graph into a risk assessment model, combining the dynamic change trend of the dynamic asset profile with the abnormal propagation path in the asset semantic association network, quantitatively scoring the compliance risk, value decay risk, and security threat of each digital asset, and generating an asset risk situation map; based on the asset risk situation map, generating asset management strategies for assets of different risk levels by a strategy generation model, and issuing and executing them through a unified management interface; collecting asset status change data after the execution of the asset management strategy as a feedback signal to incrementally learn and update the feature extraction model and the risk assessment model. By deeply integrating dynamic asset profiling with knowledge graphs, the implicit relationships between digital assets are visualized and perceived, solving the problems of isolated asset management, delayed risk assessment, and rigid strategies in existing solutions. With the help of a closed-loop feedback mechanism, the accuracy of risk identification and the adaptability of strategies are continuously improved, significantly enhancing the intelligence level and security protection capabilities of digital asset management. Attached Figure Description

[0018] Figure 1 This is a flowchart of an artificial intelligence-based digital asset management method provided in one embodiment of the present invention; Figure 2 This is a schematic block diagram of an artificial intelligence-based digital asset management system provided in one embodiment of the present invention. Detailed Implementation

[0019] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0020] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0021] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0022] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0023] The following reference Figures 1 to 2 This invention describes an artificial intelligence-based digital asset management method and system provided by some embodiments of the present invention.

[0024] like Figure 1 As shown, one embodiment of the present invention provides a digital asset management method based on artificial intelligence, comprising: By using a multimodal feature extraction model, feature vectors are extracted from the dimensions of content semantics, behavioral trajectory, association relationship and life cycle status of digital assets, and a dynamic asset profile that can be updated in real time as the asset status changes. Using the dynamic asset profile as nodes, an asset semantic association network is constructed based on the semantic similarity, ownership relationship and business dependency relationship between assets, and the asset semantic association network is stored in a knowledge graph, so that the implicit association between cross-type and cross-domain digital assets can be expressed explicitly. The knowledge graph is input into the risk assessment model, and combined with the dynamic change trend of the dynamic asset profile and the abnormal propagation path in the asset semantic association network, the compliance risk, value decay risk and security threat of each digital asset are quantitatively scored to generate an asset risk situation map. Based on the asset risk profile, the strategy generation model generates asset management strategies (including automatic asset matching access control, encryption protection, backup migration or deregistration disposal, etc.) for assets of different risk levels, and issues them for execution through a unified management interface to achieve closed-loop management of the entire asset lifecycle. Data on changes in asset status after the implementation of asset management strategies are collected as feedback signals to incrementally learn and update the feature extraction model and risk assessment model, enabling the management strategy to continuously optimize itself as the asset environment evolves.

[0025] The technical solution adopted in this embodiment achieves visualized perception of implicit associations of digital assets through the deep integration of dynamic asset profiling and knowledge graphs, solving the problems of isolated asset management, delayed risk assessment, and rigid strategies in existing solutions; with the help of a closed-loop feedback mechanism, the accuracy of risk identification and the adaptability of strategies are continuously improved, significantly enhancing the intelligence level and security protection capabilities of digital asset management.

[0026] In some possible embodiments of the present invention, the step of using a multimodal feature extraction model to extract feature vectors from the dimensions of content semantics, behavioral trajectory, association relationship, and lifecycle state of digital assets, and constructing a dynamic asset profile that can be updated in real time as the asset state changes, includes: The collected digital assets are automatically identified and classified according to their data modalities, and text, image, structured data and behavior log assets are respectively diverted to the corresponding single-modal feature extraction sub-models in the multimodal feature extraction model; Each single-modal feature extraction sub-model extracts content semantic vectors, behavioral trajectory sequence vectors, relational topology vectors, and lifecycle state vectors from the corresponding assets to form local feature vectors for each modality. The local feature vectors of each modality are input into the cross-modal attention fusion module, and adaptive weighted fusion is performed based on the correlation weights between features of different modalities to output a unified multimodal comprehensive feature vector; Using multimodal integrated feature vectors as the core and combining asset metadata, an initial dynamic profile of the digital asset is generated, and the profile is stored in the profile library in an incrementally updatable structure. When any modal feature of an asset is detected to change, the corresponding single-modal sub-model is triggered to re-extract local features, and the corresponding portrait in the portrait library is locally refreshed through the cross-modal attention fusion module to maintain the continuous consistency between the dynamic portrait and the real state of the asset.

[0027] In this embodiment, the multi-dimensional features of digital assets are accurately characterized by modal splitting extraction and cross-modal attention fusion; the incremental triggering profile update mechanism ensures that the profile always reflects the latest real state of the assets, improving the timeliness and accuracy of subsequent risk assessment.

[0028] In some possible embodiments of the present invention, the step of constructing an asset semantic association network based on the dynamic asset profile as nodes, according to the semantic similarity, ownership relationship, and business dependency relationship between assets, and storing the asset semantic association network in a knowledge graph includes: Three types of relationships between assets are predefined, including semantic similarity relationships, ownership relationships, and business dependency relationships. Based on these three types of relationships and their constraint rules, a digital asset domain ontology is constructed as a unified semantic framework for subsequent generation of relationship networks and storage of knowledge graphs. Using the multimodal comprehensive feature vector of each dynamic asset profile as input, the semantic similarity calculation model is used to measure the similarity between each profile. When the similarity exceeds the adaptive threshold, semantic similarity edges are generated between the corresponding profile nodes. Based on three types of relationships, the ownership relationship and business dependency relationship between assets are automatically extracted from asset metadata and business system configuration information, and converted into directed relationship edges of the corresponding types to obtain ownership relationship edges and business dependency relationship edges. By integrating semantic similarity edges with attribution and business dependency edges, and using dynamic asset profiles as common nodes, an asset semantic association network containing multiple types of heterogeneous edges is constructed. Based on the domain ontology, nodes and various relation edges in the asset semantic association network are mapped to entity and attribute relation triples in the knowledge graph and persistently stored in the graph database (so that implicit associations between cross-type and cross-domain assets can be expressed in the form of explicit triples in a structured manner). When any dynamic asset profile in the profile library is triggered to be updated, the corresponding entity node in the knowledge graph is used as the starting point for the update. The semantic similarity calculation, relation edge extraction, heterogeneous network fusion and triple mapping storage operations involving the entity node are re-executed to incrementally update the entity node and its related relation triples in the knowledge graph, so as to maintain the continuous consistency between the knowledge graph and the real association state of the asset.

[0029] In this embodiment, by integrating a unified semantic framework of domain ontology with heterogeneous edges of multiple types, the implicit associations of cross-type and cross-domain digital assets are made explicit and structured. The incremental synchronization mechanism ensures that the knowledge graph dynamically reflects the current status of asset associations, providing an accurate and reliable graph structure foundation for subsequent risk propagation path analysis.

[0030] In some possible embodiments of the present invention, the step of inputting the knowledge graph into a risk assessment model, combining the dynamic changing trends of the dynamic asset profile with the abnormal propagation paths in the asset semantic association network, and quantitatively scoring the compliance risks, value decay risks, and security threats of each digital asset to generate an asset risk situation map includes: Time-series modeling is performed on the historical update sequence of each dynamic asset portrait in the portrait library to extract the portrait change trend features of each dynamic asset portrait in terms of content semantics, behavior trajectory, relationship and life cycle status. Using a knowledge graph as input and the entity nodes corresponding to each digital asset in the knowledge graph as the operation objects, the features of the neighboring entity nodes of each entity node are aggregated through a graph neural network. The aggregated node representation is compared with the baseline representation under the normal association mode to identify abnormal entity nodes that deviate from the normal association mode. The abnormal propagation path of abnormal entity nodes spreading to neighboring entity nodes along the affiliation relationship edge and the business dependency relationship edge is tracked to locate the potential links of risk spread. By inputting the characteristics of profile change trends and abnormal propagation paths into the risk assessment model, each digital asset is independently quantitatively scored from three dimensions: compliance risk, value decay risk, and security threat, resulting in a three-dimensional risk score vector for each asset. Based on the characteristics of abnormal propagation paths, the three-dimensional risk score vectors of related assets in the propagation chain are transmitted and corrected so that the risk scores reflect the superposition or amplification effect of mutual influence between assets. Using a knowledge graph as the topological framework, the corrected three-dimensional risk score vectors of each asset are used as node attributes, and the anomaly propagation path is used as the risk transmission edge attribute. These are overlaid and rendered onto the knowledge graph to generate an asset risk situation map that intuitively presents the risk level of each asset and the risk transmission relationship. Continuously monitor changes in the profile database and knowledge graph. When any dynamic asset profile or knowledge graph relationship is updated, the asset risk situation graph is incrementally refreshed to achieve continuous and dynamic perception of the risk situation.

[0031] In this embodiment, by integrating the temporal change trend of asset profiles with the abnormal propagation path of knowledge graph, a multi-dimensional collaborative assessment of three types of risks—compliance, value decay, and security threats—is achieved. The associated risk transmission correction mechanism effectively captures the superposition effect of cross-asset risks, significantly improving the comprehensiveness of risk identification and the accuracy of assessment results.

[0032] In some possible embodiments of the present invention, the step of generating asset management strategies for assets of different risk levels by a strategy generation model based on the asset risk profile, and issuing and executing these strategies through a unified management interface, includes: Based on the three-dimensional risk scoring vector of each asset in the asset risk situation map, and according to the preset risk level classification rules, all digital assets are classified into four levels: high risk, medium risk, low risk and normal, and the archiving results are written into the asset risk level ledger. Using the risk level and three-dimensional risk score vector of each asset as input, a set of candidate management strategies that match the current risk characteristics is retrieved from a pre-built strategy candidate library. The candidate strategies cover types such as access control tightening, encryption protection upgrade, backup migration and deregistration. The candidate management strategy set, together with the risk transmission edge attributes of the corresponding assets in the asset risk situation map, are input into the strategy generation model. The strategy generation model comprehensively evaluates the suppression effect of each candidate strategy on the risk of the current asset and its related assets, and outputs the optimal management strategy for each asset. Perform global conflict detection on the optimal management strategy for all assets, identify strategy execution conflicts caused by business dependencies between assets, adjust conflicting strategies through priority coordination rules, and output a final set of conflict-free strategy executions. The final policy execution set is arranged according to asset ownership domain and policy type, and execution instructions are issued to the corresponding asset management system through a unified management interface to trigger the automated execution of access control policy updates, encryption parameter configuration, backup migration tasks or deregistration processes. Collect the results and status of each strategy after execution, write the results back to the asset risk level ledger, update the life cycle management record of the corresponding asset, and form a closed-loop management of the entire life cycle from risk perception, strategy decision-making to execution and write-off.

[0033] In this embodiment, the strategy generation model is combined with risk transmission attributes to make optimized decisions, and a global conflict detection and coordination mechanism is introduced to realize the automatic generation and execution of differentiated and conflict-free strategies for assets with different risk levels; the closed-loop write-back of execution results ensures the integrity and traceability of asset lifecycle management records.

[0034] In some possible embodiments of the present invention, the step of collecting asset status change data after the execution of the asset management strategy as a feedback signal to incrementally learn and update the feature extraction model and the risk assessment model includes: Using the dynamic asset profile and asset risk profile map before the strategy is issued and executed as a baseline snapshot, the profile features and risk scores of the corresponding assets are re-collected after the strategy is executed, and the state difference before and after execution is calculated to obtain a structured feedback signal that reflects the actual effect of the strategy. The feedback signals are filtered to remove abnormal state change samples caused by external interference, and retain valid feedback samples that can truly reflect the causal effect of the strategy. Based on the execution result write-back records in the strategy execution result ledger, the valid samples are automatically labeled with positive and negative effect tags. Using the changes in asset profile features in the effective feedback samples as incremental training data, the multimodal feature extraction sub-model and the cross-modal attention fusion module are incrementally learned, so that the feature extraction model can continuously adapt to the evolution of asset content, behavior and relationships. Using the risk score changes and abnormal propagation path correction information in the effective feedback samples as incremental training data, the graph neural network and risk assessment model are incrementally learned to correct the model's assessment bias of compliance risk, value decay risk and security threat. Based on positive and negative effect labels, the applicable conditions and priority weights of corresponding strategies in the strategy candidate library are adjusted, and the strategy generation model is incrementally fine-tuned with effective feedback samples, so that the strategy decision-making ability can be continuously improved as the asset environment evolves. After each round of incremental learning, the updated models are evaluated using the retained validation sample set. When the evaluation metric fails to reach the preset improvement threshold, a retraining process with an expanded sample range is triggered to ensure the effectiveness of incremental learning and prevent performance degradation of the model due to low-quality feedback.

[0035] In this embodiment, a high-quality feedback signal is constructed by differentiating the states before and after strategy execution, which drives the collaborative incremental update of feature extraction, risk assessment and strategy generation models, realizing the continuous self-optimization of the management system as the asset environment evolves; the evaluation triggering mechanism effectively prevents model performance degradation caused by low-quality feedback and ensures the long-term stable operation of the system.

[0036] In some possible embodiments of the present invention, when the local feature vectors of each modality are input into the cross-modal attention fusion module, and adaptive weighted fusion is performed based on the correlation weights between features of different modalities to output a unified multimodal comprehensive feature vector, the fusion weights of each modality are determined in the following way: For digital asset a, the adaptive fusion weights of its p-th mode Calculated by the following formula:

[0037] in, The query vector is obtained by linear projection of the p-th modal feature vector of asset a; For the corresponding key vector; Let M be the dimension of the feature space of the p-th modality, used to scale the dot product to prevent gradient vanishing; M is the total number of modalities participating in the fusion. , which is the learnable reliability prior weight of the p-th mode, reflecting the overall reliability of the mode in characterizing the asset state during the historical fusion process; The real-time information integrity score for the p-th modality of asset a is determined by the missing rate and noise level of the currently collected data for that modality. When a modality has severe data missing or high noise, a score is awarded. A lower value is chosen to reduce the fusion weight of this mode and prevent low-quality modes from interfering with the accuracy of the comprehensive feature vector. Final multimodal integrated feature vector We obtain it from the following formula:

[0038] in, Let be the value vector obtained by linear projection of the p-th modal eigenvector of asset a; The cross-modal residual compensation vector for asset a is obtained by linear transformation of the weighted residuals of each modality feature vector. It is used to compensate for feature loss caused by intermodal information redundancy or alignment deviation during the attention weighted summation process.

[0039] In some possible embodiments of the present invention, when measuring the similarity of images pairwise using a semantic similarity calculation model, the cross-domain semantic similarity between any two digital assets a and b is... Calculated by the following formula:

[0040] in, These are the multimodal integrated feature vectors of assets a and b, respectively. The cosine similarity function measures the degree of similarity between two assets in the semantic content dimension. These are the category code vectors for assets a and b in the business classification system, respectively. For Euclidean distance, This is the smoothing coefficient for the class distance; Let be the shortest number of hops between assets a and b in the knowledge graph. The threshold for the number of hops. For indicator functions; These are the sets of operation types that assets a and b have appeared in the historical behavior trajectory sequence, respectively. Jaccard similarity for the historical operation types of two assets is used to measure the degree of commonality between the two assets in the behavioral pattern dimension. These are the coded values ​​representing the current lifecycle stages of assets a and b, respectively. For life cycle stage gap, As a smoothing coefficient for the life cycle stage gap, this exponential decay term gives assets with similar life cycle stages a higher behavioral similarity bonus, avoiding the incorrect classification of assets in different life cycle stages as strongly correlated due to the overlap of historical behavior sets. The learnable balance coefficients for the four similarity components satisfy... , .

[0041] In some possible embodiments of the present invention, when performing three-dimensional risk quantification scoring on various digital assets, the security threat score of asset a is... Calculated by the following formula:

[0042] in, It is a set of neighboring assets in the knowledge graph that have business dependencies or ownership relationships with asset a; The probability of a basic security threat to the neighboring asset n itself; The edge weights for transmitting security threats from asset n to asset a are determined by the degree of anomaly of the corresponding knowledge graph relation edges; This is the propagation path protection coefficient between asset n and asset a, with a value range of (0,1]. It reflects the strength of the existing access control and encryption protection measures between the two assets in suppressing threat propagation. The more comprehensive the protection measures, the stronger the protection. The smaller the value, the lower the actual threat contribution of this transmission path; The vulnerability coefficient of asset a itself is determined by the aging degree of the lifecycle state vector in the asset profile and the completeness of the security configuration. A function representing the magnitude of change in the profile of asset a. This represents the change in the current image feature vector relative to the previous time step. for A subset of neighboring assets whose current security threat score exceeds the high-risk threshold. Its quantity; Let be the sensitivity coefficient of asset a to the number of high-risk neighboring assets. The logarithmic term is used to characterize the nonlinear amplification effect of the increase in the number of high-risk neighboring assets on the security threat of asset a. The logarithmic form is used to prevent the score from expanding infinitely when there are too many high-risk neighboring assets.

[0043] In some possible embodiments of the present invention, when the strategy generation model makes optimization decisions on each candidate management strategy, asset a is in the candidate strategy set. Optimal Strategy The choice is determined by the following formula:

[0044] in, Let t be the expected suppression return of strategy t on the three-dimensional risk score vector of asset a itself; The expected cascading suppression of returns on neighboring related assets after the execution of strategy t via risk transmission edge; The resource cost for executing strategy t is calculated by comprehensively considering storage, computing, and business interruption costs. The expected conflict penalty for strategy t that conflicts with other decided strategies in the current asset risk level ledger; The historical success rate weighted return of strategy t is obtained by the weighted product of the historical success rate of strategy t on assets of the same risk level and the corresponding historical suppression return in the strategy candidate library. It is used to guide strategy decision-making to prioritize well-validated high success rate strategies. Let strategy t be the expected duration of interruption caused by normal business access to asset a. This is the service interruption tolerance threshold. The smoothing coefficient for the interruption duration penalty is used when the interruption duration exceeds a threshold. The strategy of imposing a non-linear, incremental penalty on the score to strongly constrain excessive impact on business continuity was selected. The indicator function is used to activate this penalty term; These are the learnable balancing weight coefficients for each item.

[0045] In some possible embodiments of the present invention, when evaluating the incremental learning effect, a comprehensive model optimization effect score is calculated after each round of incremental learning. Calculated by the following formula:

[0046] in, The set of models participating in the evaluation includes multimodal feature extraction models, risk assessment models, and strategy generation models; These are the performance metrics of model m before and after incremental learning on the validation sample set, respectively. The importance weight of model m in the overall management chain; To prevent the smoothing constant from having a denominator of zero; This is the performance degradation penalty coefficient; This is an indicator function used to activate the degradation penalty term; Let be the consistency coefficient between the current validation sample set and the training sample distribution of model m, with a value range of (0,1]. When the difference between the validation sample distribution and the training sample distribution is large... A lower value is used to discount the virtual performance improvement caused by distribution shift, so as to prevent misjudgment of the incremental learning effect in the distribution shift scenario. This represents the number of effective feedback samples used by model m in this round of incremental learning. The sample size saturation smoothing coefficient is the exponential saturation term. This is used to characterize the non-linear moderating effect of the number of effective samples on the credibility of the score: when the number of samples is small, the credibility of the score is low, and its contribution to the overall score is compressed; when the number of samples is sufficient, the credibility of the score tends to saturate, preventing the score from being artificially inflated by simply piling up the number of samples.

[0047] Please refer to Figure 2 Another embodiment of the present invention provides an artificial intelligence-based digital asset management system for executing an artificial intelligence-based digital asset management method, comprising: a server and an asset management terminal; The server is configured as follows: By using a multimodal feature extraction model, feature vectors are extracted from the dimensions of content semantics, behavioral trajectory, association relationship and life cycle status of digital assets, and a dynamic asset profile that can be updated in real time as the asset status changes. Using the dynamic asset profile as nodes, an asset semantic association network is constructed based on the semantic similarity, ownership relationship and business dependency relationship between assets, and the asset semantic association network is stored in a knowledge graph; The knowledge graph is input into the risk assessment model, and combined with the dynamic change trend of the dynamic asset profile and the abnormal propagation path in the asset semantic association network, the compliance risk, value decay risk and security threat of each digital asset are quantitatively scored to generate an asset risk situation map. Based on the asset risk profile, the strategy generation model generates asset management strategies for assets with different risk levels, and then issues and executes them through a unified management interface. Data on changes in asset status after the implementation of asset management strategies are collected as feedback signals to incrementally learn and update the feature extraction model and risk assessment model.

[0048] It should be known that, Figure 2 The block diagram of the AI-based digital asset management system shown is for illustrative purposes only, and the number of modules shown does not limit the scope of protection of this invention. The AI-based digital asset management system provided in this embodiment can be used to execute various embodiments of the corresponding AI-based digital asset management methods. For specific implementation details, please refer to the descriptions of the respective method embodiments, which will not be repeated here.

[0049] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0050] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0051] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical or other forms.

[0052] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0053] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0054] If the integrated units described above are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0055] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0056] The embodiments of this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

[0057] While the present invention has been disclosed above, it is not limited thereto. Any person skilled in the art can easily conceive of variations or substitutions without departing from the spirit and scope of the present invention, and various modifications and alterations can be made, including combinations of the different functions and implementation steps described above, as well as software and hardware implementation methods, all of which are within the protection scope of the present invention.

Claims

1. A digital asset management method based on artificial intelligence, characterized in that, include: By using a multimodal feature extraction model, feature vectors are extracted from the dimensions of content semantics, behavioral trajectory, association relationship and life cycle status of digital assets, and a dynamic asset profile that can be updated in real time as the asset status changes. Using the dynamic asset profile as nodes, an asset semantic association network is constructed based on the semantic similarity, ownership relationship and business dependency relationship between assets, and the asset semantic association network is stored in a knowledge graph; The knowledge graph is input into the risk assessment model, and combined with the dynamic change trend of the dynamic asset profile and the abnormal propagation path in the asset semantic association network, the compliance risk, value decay risk and security threat of each digital asset are quantitatively scored to generate an asset risk situation map. Based on the asset risk profile, the strategy generation model generates asset management strategies for assets with different risk levels, and then issues and executes them through a unified management interface. Data on changes in asset status after the implementation of asset management strategies are collected as feedback signals to incrementally learn and update the feature extraction model and risk assessment model.

2. The artificial intelligence-based digital asset management method according to claim 1, characterized in that, The step of using a multimodal feature extraction model to extract feature vectors from the dimensions of content semantics, behavioral trajectory, association relationship, and lifecycle state of digital assets, and constructing a dynamic asset profile that can be updated in real time as the asset state changes, includes: The collected digital assets are automatically identified and classified according to their data modalities, and text, image, structured data and behavior log assets are respectively diverted to the corresponding single-modal feature extraction sub-models in the multimodal feature extraction model; Each single-modal feature extraction sub-model extracts content semantic vectors, behavioral trajectory sequence vectors, relational topology vectors, and lifecycle state vectors from the corresponding assets to form local feature vectors for each modality. The local feature vectors of each modality are input into the cross-modal attention fusion module, and adaptive weighted fusion is performed based on the correlation weights between features of different modalities to output a unified multimodal comprehensive feature vector; Using multimodal integrated feature vectors as the core and combining asset metadata, an initial dynamic profile of the digital asset is generated, and the profile is stored in the profile library in an incrementally updatable structure. When any modal feature of an asset is detected to change, the corresponding single-modal sub-model is triggered to re-extract local features, and the corresponding portrait in the portrait library is locally refreshed through the cross-modal attention fusion module.

3. The artificial intelligence-based digital asset management method according to claim 2, characterized in that, The step of constructing an asset semantic association network based on the dynamic asset profile as nodes, according to the semantic similarity, ownership relationship, and business dependency relationship between assets, and storing the asset semantic association network in a knowledge graph includes: Three types of relationships between assets are predefined, including semantic similarity relationships, ownership relationships, and business dependency relationships. Based on these three types of relationships and their constraint rules, a digital asset domain ontology is constructed as a unified semantic framework for subsequent generation of relationship networks and storage of knowledge graphs. Using the multimodal comprehensive feature vector of each dynamic asset profile as input, the semantic similarity calculation model is used to measure the similarity between each profile. When the similarity exceeds the adaptive threshold, semantic similarity edges are generated between the corresponding profile nodes. Based on three types of relationships, the ownership relationship and business dependency relationship between assets are automatically extracted from asset metadata and business system configuration information, and converted into directed relationship edges of the corresponding types to obtain ownership relationship edges and business dependency relationship edges. By integrating semantic similarity edges with attribution and business dependency edges, and using dynamic asset profiles as common nodes, an asset semantic association network containing multiple types of heterogeneous edges is constructed. Based on the domain ontology, the nodes and various relation edges in the asset semantic association network are mapped to entity and attribute relation triples in the knowledge graph and persistently stored in the graph database. When any dynamic asset portrait in the portrait library is triggered to be updated, the entity node corresponding to this dynamic asset portrait in the knowledge graph is taken as the update starting point. The semantic similarity calculation, relation edge extraction, heterogeneous network fusion and triple mapping storage operations involving the entity node are re-executed to incrementally update the entity node and its related relation triples in the knowledge graph.

4. The artificial intelligence-based digital asset management method according to claim 3, characterized in that, The step of inputting the knowledge graph into the risk assessment model, combining the dynamic changes of the dynamic asset profile with the abnormal propagation path in the asset semantic association network, to quantify and score the compliance risk, value decay risk, and security threat of each digital asset, and generate an asset risk situation map, includes: Time-series modeling is performed on the historical update sequence of each dynamic asset portrait in the portrait library to extract the portrait change trend features of each dynamic asset portrait in terms of content semantics, behavior trajectory, relationship and life cycle status. Using a knowledge graph as input and the entity nodes corresponding to each digital asset in the knowledge graph as the operation objects, the features of the neighboring entity nodes of each entity node are aggregated through a graph neural network. The aggregated node representation is compared with the baseline representation under the normal association mode to identify abnormal entity nodes that deviate from the normal association mode. The abnormal propagation path of abnormal entity nodes spreading to neighboring entity nodes along the affiliation relationship edge and the business dependency relationship edge is tracked to locate the potential links of risk spread. By inputting the characteristics of profile change trends and abnormal propagation paths into the risk assessment model, each digital asset is independently quantitatively scored from three dimensions: compliance risk, value decay risk, and security threat, resulting in a three-dimensional risk score vector for each asset. Based on the characteristics of abnormal propagation paths, the three-dimensional risk score vectors of related assets in the propagation chain are transmitted and corrected so that the risk scores reflect the superposition or amplification effect of mutual influence between assets. Using a knowledge graph as the topological framework, the corrected three-dimensional risk score vectors of each asset are used as node attributes, and the anomaly propagation path is used as the risk transmission edge attribute. These are overlaid and rendered onto the knowledge graph to generate an asset risk situation map that intuitively presents the risk level of each asset and the risk transmission relationship. Continuously monitor changes in the profile database and knowledge graph. When any dynamic asset profile or knowledge graph relationship is updated, the asset risk situation graph is incrementally refreshed to achieve continuous and dynamic perception of the risk situation.

5. The artificial intelligence-based digital asset management method according to claim 4, characterized in that, The step of generating asset management strategies for assets with different risk levels based on the asset risk profile and issuing them for execution through a unified management interface includes: Based on the three-dimensional risk scoring vector of each asset in the asset risk situation map, and according to the preset risk level classification rules, all digital assets are classified into four levels: high risk, medium risk, low risk and normal, and the archiving results are written into the asset risk level ledger. Using the risk level and three-dimensional risk score vector of each asset as input, a set of candidate management strategies that match the current risk characteristics is retrieved from a pre-built strategy candidate library. The candidate strategies cover types such as access control tightening, encryption protection upgrade, backup migration and deregistration. The candidate management strategy set, together with the risk transmission edge attributes of the corresponding assets in the asset risk situation map, are input into the strategy generation model. The strategy generation model comprehensively evaluates the suppression effect of each candidate strategy on the risk of the current asset and its related assets, and outputs the optimal management strategy for each asset. Perform global conflict detection on the optimal management strategy for all assets, identify strategy execution conflicts caused by business dependencies between assets, adjust conflicting strategies through priority coordination rules, and output a final set of conflict-free strategy executions. The final policy execution set is arranged according to asset ownership domain and policy type, and execution instructions are issued to the corresponding asset management system through a unified management interface to trigger the automated execution of access control policy updates, encryption parameter configuration, backup migration tasks or deregistration processes. Collect the results and status of each strategy after execution, write the results back to the asset risk level ledger, update the life cycle management record of the corresponding asset, and form a closed-loop management of the entire life cycle from risk perception, strategy decision-making to execution and write-off.

6. The artificial intelligence-based digital asset management method according to claim 5, characterized in that, The step of using data on changes in asset status after the execution of the asset management strategy as feedback signals to incrementally learn and update the feature extraction model and the risk assessment model includes: Using the dynamic asset profile and asset risk profile map before the strategy is issued and executed as a baseline snapshot, the profile features and risk scores of the corresponding assets are re-collected after the strategy is executed, and the state difference before and after execution is calculated to obtain a structured feedback signal that reflects the actual effect of the strategy. The feedback signals are filtered to remove abnormal state change samples caused by external interference, and retain valid feedback samples that can truly reflect the causal effect of the strategy. Based on the execution result write-back records in the strategy execution result ledger, the valid samples are automatically labeled with positive and negative effect tags. Using the changes in asset profile features in the effective feedback samples as incremental training data, the multimodal feature extraction sub-model and the cross-modal attention fusion module are incrementally learned, so that the feature extraction model can continuously adapt to the evolution of asset content, behavior and relationships. Using the risk score changes and abnormal propagation path correction information in the effective feedback samples as incremental training data, the graph neural network and risk assessment model are incrementally learned to correct the model's assessment bias of compliance risk, value decay risk and security threat. Based on positive and negative effect labels, the applicable conditions and priority weights of corresponding strategies in the strategy candidate library are adjusted, and the strategy generation model is incrementally fine-tuned with effective feedback samples, so that the strategy decision-making ability can be continuously improved as the asset environment evolves. After each round of incremental learning, the updated models are evaluated using the retained validation sample set. When the evaluation metric fails to reach the preset improvement threshold, a retraining process with an expanded sample range is triggered to ensure the effectiveness of incremental learning and prevent performance degradation of the model due to low-quality feedback.

7. The artificial intelligence-based digital asset management method according to claim 6, characterized in that, When measuring the pairwise similarity of profiles using a semantic similarity calculation model, the cross-domain semantic similarity between any two digital assets a and b is... Calculated by the following formula: in, These are the multimodal integrated feature vectors of assets a and b, respectively. The cosine similarity function measures the degree of similarity between two assets in the semantic content dimension. These are the category code vectors for assets a and b in the business classification system, respectively. For Euclidean distance, This is the smoothing coefficient for the class distance; Let be the shortest number of hops between assets a and b in the knowledge graph. The threshold for the number of hops. For indicator functions; These are the sets of operation types that assets a and b have appeared in the historical behavior trajectory sequence, respectively. Jaccard similarity for the historical operation types of two assets is used to measure the degree of commonality between the two assets in the behavioral pattern dimension. These are the coded values ​​representing the current lifecycle stages of assets a and b, respectively. For life cycle stage gap, As a smoothing coefficient for the life cycle stage gap, this exponential decay term gives assets with similar life cycle stages a higher behavioral similarity bonus, avoiding the incorrect classification of assets in different life cycle stages as strongly correlated due to the overlap of historical behavior sets. The learnable balance coefficients for the four similarity components satisfy... , .

8. The artificial intelligence-based digital asset management method according to claim 7, characterized in that, When performing three-dimensional risk quantification scoring on various digital assets, the security threat score of asset a is... Calculated by the following formula: in, It is a set of neighboring assets in the knowledge graph that have business dependencies or ownership relationships with asset a; The probability of a basic security threat to the neighboring asset n itself; The edge weights for transmitting security threats from asset n to asset a are determined by the degree of anomaly of the corresponding knowledge graph relation edges; This is the propagation path protection coefficient between asset n and asset a, with a value range of (0,1]. It reflects the strength of the existing access control and encryption protection measures between the two assets in suppressing threat propagation. The more comprehensive the protection measures, the stronger the protection. The smaller the value, the lower the actual threat contribution of this transmission path; The vulnerability coefficient of asset a itself is determined by the aging degree of the lifecycle state vector in the asset profile and the completeness of the security configuration. A function representing the magnitude of change in the profile of asset a. This represents the change in the current image feature vector relative to the previous time step. for A subset of neighboring assets whose current security threat score exceeds the high-risk threshold. Its quantity; Let be the sensitivity coefficient of asset a to the number of high-risk neighboring assets. The logarithmic term is used to characterize the nonlinear amplification effect of the increase in the number of high-risk neighboring assets on the security threat of asset a. The logarithmic form is used to prevent the score from expanding infinitely when there are too many high-risk neighboring assets.

9. The artificial intelligence-based digital asset management method according to claim 8, characterized in that, When the strategy generation model makes optimization decisions on each candidate management strategy, asset a is in the candidate strategy set. Optimal Strategy The choice is determined by the following formula: in, Let t be the expected suppression return of strategy t on the three-dimensional risk score vector of asset a itself; The expected cascading suppression of returns on neighboring related assets after the execution of strategy t via risk transmission edge; The resource cost for executing strategy t is calculated by comprehensively considering storage, computing, and business interruption costs. The expected conflict penalty for strategy t that conflicts with other decided strategies in the current asset risk level ledger; The historical success rate weighted return of strategy t is obtained by the weighted product of the historical success rate of strategy t on assets of the same risk level and the corresponding historical suppression return in the strategy candidate library. It is used to guide strategy decision-making to prioritize well-validated high success rate strategies. Let strategy t be the expected duration of interruption caused by normal business access to asset a. This is the service interruption tolerance threshold. The smoothing coefficient for the interruption duration penalty is used when the interruption duration exceeds a threshold. The strategy of imposing a non-linear, incremental penalty on the score to strongly constrain excessive impact on business continuity was selected. The indicator function is used to activate this penalty term; These are the learnable balancing weight coefficients for each item.

10. An artificial intelligence-based digital asset management system, used to execute the artificial intelligence-based digital asset management method as described in any one of claims 1 to 9, characterized in that, include: Servers and asset management terminals; The server is configured as follows: By using a multimodal feature extraction model, feature vectors are extracted from the dimensions of content semantics, behavioral trajectory, association relationship and life cycle status of digital assets, and a dynamic asset profile that can be updated in real time as the asset status changes. Using the dynamic asset profile as nodes, an asset semantic association network is constructed based on the semantic similarity, ownership relationship and business dependency relationship between assets, and the asset semantic association network is stored in a knowledge graph; The knowledge graph is input into the risk assessment model, and combined with the dynamic change trend of the dynamic asset profile and the abnormal propagation path in the asset semantic association network, the compliance risk, value decay risk and security threat of each digital asset are quantitatively scored to generate an asset risk situation map. Based on the asset risk profile, the strategy generation model generates asset management strategies for assets with different risk levels, and then issues and executes them through a unified management interface. Data on changes in asset status after the implementation of asset management strategies are collected as feedback signals to incrementally learn and update the feature extraction model and risk assessment model.

Images