Digital asset lifecycle management system and method

Abstract

The present application relates to the technical field of digital asset management, in particular to a digital asset whole life cycle management system and method, comprising: obtaining metadata, flow record and operation log set of target digital asset, using improved graph embedding algorithm combining semantic association and flow topology optimization to generate asset feature embedding vector for the metadata, extracting flow record structure mode to generate multi-level asset flow graph, analyzing and encoding operation log to generate asset operation behavior vector sequence, inputting the three types of data into a dynamic fusion network to obtain a comprehensive state tensor, analyzing the tensor through an asset value evaluation network with attention mechanism, outputting a multi-life stage value density distribution, and then obtaining an adaptive management strategy set covering the whole process from creation to destruction of the digital asset through a strategy generation model. The present application realizes adaptive management of the whole life cycle of the digital asset through multi-dimensional data fusion and fine value evaluation, and optimizes the precision of digital asset management and control.

Description

Technical Field

[0001] This invention relates to the field of digital asset management technology, and in particular to a digital asset lifecycle management system and method. Background Technology

[0002] Existing technologies related to the full lifecycle management of digital assets typically process the metadata set, circulation record set, and operation log set of digital assets independently. Metadata processing uses traditional graph embedding algorithms to extract features, circulation records only perform basic node information organization, and operation logs only retain the original operation information. Digital asset valuation mostly adopts static statistical analysis methods, and the corresponding management strategies are mostly fixed process settings, without being adjusted in conjunction with the dynamic changes throughout the entire lifecycle of digital assets.

[0003] Traditional technologies fail to optimize embedding algorithms based on the semantic relationships and circulation topology between digital assets. This results in the inability to achieve unified dynamic fusion processing of multiple data types, making it difficult to generate comprehensive data representing the entire lifecycle of digital assets. Furthermore, attention mechanisms cannot be used to analyze the value density distribution across multiple asset lifecycle stages, and management strategies cannot cover the dynamic adaptation needs of digital assets from creation, circulation, use, archiving to destruction. Optimized graph embedding processing of metadata is needed, fusing circulation graphs and behavioral vector sequences to form a comprehensive state tensor. Attention mechanisms should then be used to perform phased value density analysis, generating a set of adaptive management strategies for the entire process, thus overcoming the limitations of existing technologies in data processing, value assessment, and strategy generation. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a digital asset lifecycle management system and method.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a digital asset lifecycle management method, comprising:

[0006] Obtain the metadata set, transfer record set, and operation log set of the target digital asset. The metadata set includes an asset content summary, asset type tags, and a list of permission definitions. The transfer record set includes the transfer time and subject information of multiple nodes. The operation log set includes access, modification, copy, and deletion operation information at multiple times.

[0007] An improved graph embedding algorithm is applied to the metadata set to generate asset feature embedding vectors. The improved graph embedding algorithm optimizes the traditional embedding algorithm based on the semantic association and circulation topology between digital assets.

[0008] The set of transfer records is processed by structural pattern extraction to generate a multi-level asset transfer graph. The set of operation logs is processed by intent parsing and behavior encoding to generate an asset operation behavior vector sequence.

[0009] The asset feature embedding vector, asset flow map, and asset operation behavior vector sequence are input into a dynamic fusion network to generate a comprehensive state tensor representing the dynamics of the entire digital asset lifecycle.

[0010] The attention-based asset valuation network is invoked to analyze the comprehensive state tensor and output the value density distribution of digital assets in multiple preset life stages.

[0011] Based on the value density distribution, an adaptive management strategy set covering the entire process of digital assets from creation, circulation, use, archiving to destruction is calculated and output through a strategy generation model.

[0012] As a further aspect of the present invention, an improved graph embedding algorithm is performed on the metadata set to generate asset feature embedding vectors, including:

[0013] Based on the asset type tags and permission definition list in the metadata set, an initial asset relationship graph is constructed with digital assets as nodes and semantic associations and permission inheritance relationships between assets as edges;

[0014] On the initial asset relationship graph, historical transfer paths between assets are added as reinforcing edges based on the set of transfer records to form an enhanced asset relationship graph;

[0015] Based on the improved graph embedding algorithm, multi-hop neighbor information aggregation is performed on the enhanced asset relationship graph. During the aggregation process, the improved graph embedding algorithm defines differentiated information transmission weights for different types of edges. The differentiated information transmission weights are jointly determined by the semantic association strength, the permission inheritance depth, and the historical flow frequency.

[0016] After completing multiple rounds of information aggregation, the feature vector corresponding to each asset node is sampled and normalized, and the asset feature embedding vector corresponding to each digital asset is output.

[0017] As a further aspect of the present invention, structural pattern extraction processing is performed on the set of transfer records to generate a multi-level asset transfer map, including:

[0018] Analyze the multi-node transfer time and main information in the set of transfer records, and construct the original transfer event sequence in chronological order;

[0019] On the original sequence of events, frequently occurring patterns of transfer are identified. These patterns are defined by multiple consecutive transfer events and the relationships between the events.

[0020] Based on the identified flow patterns, the original flow event sequence is abstracted and compressed to form a high-level flow logic diagram with flow patterns as the basic unit.

[0021] Align and associate the high-level transfer logic diagram with the low-level transfer detail diagram based on a single transfer event to jointly form the multi-level asset transfer map.

[0022] As a further aspect of the present invention, intent parsing and behavior encoding processing are performed on the operation log set to generate an asset operation behavior vector sequence, including:

[0023] For each access, modification, copy, and deletion operation in the operation log set, a pre-trained natural language understanding model is invoked to perform intent classification and obtain the operation intent label;

[0024] For each operation message, extract its operation subject, operation time, operation object identifier and operation parameters, and encode the operation intention tag and the extracted information together into a structured operation record;

[0025] The structured operation records are sorted according to the operation time to form a time-ordered sequence of operation records;

[0026] The time-ordered sequence of operation records is encoded by a time encoder, and each structured operation record is transformed into a fixed-length behavior vector. The vectors are then arranged in order to obtain the asset operation behavior vector sequence.

[0027] As a further aspect of the present invention, the asset feature embedding vector, asset transfer graph, and asset operation behavior vector sequence are input into a dynamic fusion network to generate a comprehensive state tensor representing the dynamics of the entire digital asset lifecycle, including:

[0028] The dynamic fusion network includes a graph coding branch, a vector alignment branch, and a dynamic memory module;

[0029] The graph encoding branch receives the asset transfer graph and encodes it into graph structure features;

[0030] The vector alignment branch receives the asset feature embedding vector and the asset operation behavior vector sequence;

[0031] The asset feature embedding vector and the asset operation behavior vector sequence are aligned and fused in the feature dimension by using an attention mechanism to generate an aligned behavior feature sequence.

[0032] The dynamic memory module receives the graph structure features and the aligned behavioral feature sequence, and simulates the evolution of asset state in the time dimension. At each time step, the dynamic memory module updates its internal memory state based on the current behavioral features and is constrained by the graph structure features.

[0033] After the complete time period corresponding to the asset operation behavior vector sequence ends, the final state of the dynamic memory module is read and concatenated with the graph structure features to output the comprehensive state tensor.

[0034] As a further aspect of the present invention, the step of invoking an attention-based asset valuation network to analyze the comprehensive state tensor and outputting the value density distribution of the digital asset across multiple preset life stages includes:

[0035] The asset valuation network contains multiple parallel attention heads, each corresponding to a preset life stage;

[0036] The comprehensive state tensor is input into the asset valuation network, and each attention head independently weights and focuses on the features in the comprehensive state tensor that are related to the corresponding life stage.

[0037] Each attention head outputs a focused feature representation, which is fed into a fully connected layer that maps the features to a scalar representing the value density of the preset life stage corresponding to the attention head.

[0038] The scalars output from the fully connected layer corresponding to all attention points are aggregated to form the value density distribution of the digital asset across multiple preset life stages.

[0039] As a further aspect of the present invention, the adaptive management strategy set calculated and output through a strategy generation model based on the value density distribution, covering the entire process of digital assets from creation, circulation, use, archiving to destruction, includes:

[0040] The value density distribution, the comprehensive state tensor, and the preset management objective constraints are input into the strategy generation model;

[0041] Under the constraints of the management objectives and guided by the value density distribution, the strategy generation model searches for the optimal management action for each life stage in the entire strategy space from creation to destruction.

[0042] The strategy generation model assembles the optimal management action found in each life stage into a complete management strategy path by combining the transition logic between life stages.

[0043] The management strategy path is evaluated and post-processed to transform it into an adaptive management strategy set consisting of specific management rules, triggering conditions, and execution actions.

[0044] As a further aspect of the present invention, the strategy generation model, under the constraints of the management objective and guided by the value density distribution, searches for the optimal management action for each life stage in the entire strategy space from creation to destruction, including:

[0045] Treat each life stage as a decision node and construct a decision graph for the entire process from creation to destruction;

[0046] In the full-process decision diagram, the transfer cost between nodes is defined based on the management objective constraints, and the state value of each node is defined based on the value density distribution.

[0047] An improved graph search algorithm is run on the entire process decision graph. During the search process, the improved graph search algorithm simultaneously considers the state value of the current node, the cost of transitioning to the next node, and the estimated value from the next node to the destination.

[0048] The improved graph search algorithm outputs a search path, where the selected transition edge at each decision node represents the optimal management action found in the life stage.

[0049] As a further aspect of the present invention, the method further includes:

[0050] The adaptive management strategy set is transformed into an executable atomic operation instruction stream and pushed to the target digital asset management platform to drive it to perform corresponding full lifecycle management operations on the target digital asset.

[0051] The adaptive management strategy set is transformed into an executable stream of atomic operation instructions, including:

[0052] Analyze each management rule, triggering condition, and execution action in the set of adaptive management strategies;

[0053] Each management rule is broken down into logical judgment units, each triggering condition is mapped to an event listener configuration, and each execution action is translated into a call instruction for the underlying interface of the target digital asset management platform.

[0054] Based on the logical execution order of the management strategy, the logical judgment unit, event listener configuration, and underlying interface call instructions are arranged to generate an instruction sequence with clear execution logic and dependencies;

[0055] The instruction sequence is optimized by merging operations that can be executed in parallel to form the final atomic operation instruction stream.

[0056] As a further aspect of the present invention, the present invention also includes a digital asset lifecycle management system, the system including a processor and a memory, the memory and the processor being connected, the memory being used to store programs, instructions or code, and the processor being used to run the programs, instructions or code in the memory to implement the digital asset lifecycle management method as described above.

[0057] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0058] Based on the semantic relationships and circulation topology between digital assets, traditional embedding algorithms are optimized. The metadata set is processed to generate asset feature embedding vectors, the circulation record set is processed to extract structural patterns to generate a multi-level asset circulation graph, and the operation log set is processed to perform intent parsing and behavior encoding to generate asset operation behavior vector sequences. The asset feature embedding vectors, asset circulation graph, and asset operation behavior vector sequences are input into a dynamic fusion network to generate a comprehensive state tensor. This can integrate the metadata features, circulation structure features, and operation behavior features of digital assets, fully presenting the dynamic changes of digital assets throughout their entire lifecycle. This avoids the problem of incomplete feature representation caused by single-dimensional data processing and makes the asset state data fit the actual circulation and operation scenarios of digital assets.

[0059] An attention-based asset valuation network is used to analyze the comprehensive state tensor, outputting the value density distribution of digital assets across multiple preset lifecycle stages. Based on this value density distribution, a strategy generation model is used to calculate and output a set of adaptive management strategies covering the entire process from creation, circulation, use, archiving to destruction. This approach can accurately capture the value change characteristics of digital assets at different lifecycle stages and match corresponding management strategies based on the differences in value density at each stage. This ensures that the management strategies align with the actual state of each lifecycle stage of the digital asset, avoiding the disconnect between fixed management strategies and the actual operational state of the asset. The adaptability of the management strategies is aligned with the operational rules of the entire digital asset process. Attached Figure Description

[0060] Figure 1 This is a flowchart of the digital asset lifecycle management method described in this invention;

[0061] Figure 2 A flowchart for generating embedding vectors for asset features;

[0062] Figure 3 A flowchart for generating a sequence of vectors representing asset operation behaviors. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0064] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0065] See Figure 1 This invention provides a method for managing the entire lifecycle of digital assets, the specific method including:

[0066] The process involves acquiring the metadata set, circulation record set, and operation log set of the target digital asset. The metadata set includes an asset content summary, asset type tags, and a list of permission definitions. The circulation record set includes transfer times and subject information for multiple nodes. The operation log set includes access, modification, copy, and deletion operation information at multiple points in time. An improved graph embedding algorithm is applied to the metadata set to generate asset feature embedding vectors. This improved algorithm optimizes traditional embedding algorithms based on the semantic relationships and circulation topology between digital assets. Structural pattern extraction is performed on the circulation record set to generate a multi-level asset circulation graph. Intent parsing and behavior encoding are performed on the operation log set to generate a sequence of asset operation behavior vectors. The asset feature embedding vectors, asset circulation graph, and asset operation behavior vector sequences are input into a dynamic fusion network to generate a comprehensive state tensor representing the dynamics of the digital asset throughout its entire lifecycle. An attention-based asset valuation network is then used to analyze the comprehensive state tensor, outputting the value density distribution of the digital asset across multiple preset lifecycle stages. Based on the value density distribution, a set of adaptive management strategies covering the entire process of digital assets from creation, circulation, use, archiving to destruction is calculated and output through a strategy generation model.

[0067] In one embodiment of the present invention, see [reference] Figure 2Based on the asset type tags and permission definition list in the metadata set, an initial asset relationship graph is constructed, with digital assets as nodes and semantic associations and permission inheritance relationships between assets as edges. On the initial asset relationship graph, historical transfer paths between assets are added as reinforcing edges based on the transfer record set, forming an enhanced asset relationship graph. Using an improved graph embedding algorithm, multi-hop neighbor information aggregation is performed on the enhanced asset relationship graph. During the aggregation process, the improved graph embedding algorithm defines differentiated information transfer weights for different types of edges. These differentiated information transfer weights are jointly determined by the semantic association strength, permission inheritance depth, and historical transfer frequency. After completing multiple rounds of information aggregation, the feature vector corresponding to each asset node is sampled and normalized, outputting the asset feature embedding vector corresponding to each digital asset.

[0068] In practical implementation, based on the asset type tags and permission definition list in the metadata set, an initial asset relationship graph is constructed, with digital assets as nodes and semantic associations and permission inheritance relationships between assets as edges. For example, asset type tags include categories such as "design drawings," "contract text," and "user data," and the permission definition list contains the permission settings for different departments or roles to read, write, and share digital assets. Semantic association edges and permission inheritance edges between nodes are established based on this information. On the initial asset relationship graph, historical transfer paths between assets are added as reinforcing edges based on the transfer record set, forming an enhanced asset relationship graph. For example, if the transfer record set records the transfer time and subject information of a design drawing from "Design Department A Group" to "Engineering Department B Group" and then to "External Partner C Company," then timestamped transfer edges are added between the drawing node and each subject node, and high-frequency transfer paths are marked as reinforcing edges. Based on an improved graph embedding algorithm, multi-hop neighbor information aggregation is performed on the enhanced asset relationship graph. During the aggregation process, the improved algorithm defines differentiated information transmission weights for different types of edges. These differentiated weights are jointly determined by semantic association strength, permission inheritance depth, and historical transfer frequency. For example, semantic association edge weights are calculated based on asset type similarity, permission inheritance edge weights reflect the span of permission levels, and historical transfer edge weights are positively correlated with the number of transfers. During aggregation, high-weight edges transmit more feature information. After completing multiple rounds of information aggregation, the feature vectors corresponding to each asset node are sampled and normalized, outputting the asset feature embedding vector for each digital asset. For example, mean pooling is used to reduce the dimensionality of the node features after multi-hop neighbor aggregation, and then normalization is performed to obtain a fixed-dimensional embedding vector.

[0069] In some embodiments, semantic association strength is calculated based on the co-occurrence frequency and hierarchical distance of asset type tags. For example, assets belonging to the same "R&D Documents" category have a higher semantic association strength value than assets across different categories. Permission inheritance depth is determined by the hierarchical distance between roles in the permission definition list. For example, the permission inheritance depth between directly subordinate roles is set to a smaller value, while the inheritance depth across multiple levels of roles is set to a larger value. Historical transfer frequency is directly taken from the cumulative number of transfers between two nodes in the transfer record set, and logarithmically scaled to balance the numerical range.

[0070] In practical implementation, the differentiated information transmission weights can be calculated using the following formula:

[0071]

[0072] in: Represents a node With nodes Weights are passed between edges. It is a node based on metadata set computation. With nodes The strength of semantic association, The node is determined based on the permission definition list. With nodes Depth of permission inheritance, It is a node based on the statistics of the flow record set. With nodes Frequency of historical transfer , , These are the adjustment coefficients for the corresponding terms, used to balance the influence of different factors on the weights. This formula ensures that edges with strong semantic connections, tight permission inheritance, or frequent flow occupy a higher proportion in the information aggregation process.

[0073] It is understandable that the multi-round information aggregation process adopts a graph neural network architecture. In each round of aggregation, nodes collect features from their neighboring nodes and update their own features by weighted summation based on edge weights. For example, the first round of aggregation only uses direct neighbor information, while the second round expands to two-hop neighbors, capturing global structural features through multiple rounds of iteration. In some embodiments, the sampling and normalization operations include random walk sampling and layer normalization processing. For example, a fixed-length random walk is performed starting from each node to collect the node feature sequences along the path, and then layer normalization is applied to the sequences to obtain embedding vectors with stable numerical ranges. Optionally, the dimension of the asset feature embedding vector is set to 256 dimensions, which retains sufficient distinguishing information while avoiding excessive redundancy. For example, the cosine distance between the embedding vectors of different types of assets in the vector space can effectively reflect their semantic and circulation correlation.

[0074] In one embodiment of the present invention, the transfer time and subject information of multiple nodes in the transfer record set are parsed, and an original transfer event sequence is constructed in chronological order. Frequently occurring transfer patterns are identified in the original transfer event sequence. These patterns are defined by multiple consecutively occurring transfer events and the subject relationships between them. Based on the identified transfer patterns, the original transfer event sequence is abstracted and compressed to form a high-level transfer logic diagram with transfer patterns as the basic unit. The high-level transfer logic diagram is aligned and associated with the low-level transfer detail diagram based on individual transfer events to jointly construct a multi-level asset transfer map. (See also...) Figure 3 For each access, modification, copy, and deletion operation in the operation log set, a pre-trained natural language understanding model is used for intent classification to obtain operation intent labels. For each operation, the operation subject, operation time, operation object identifier, and operation parameters are extracted, and the operation intent labels and extracted information are jointly encoded into a structured operation record. The structured operation records are sorted by operation time to form a time-ordered sequence of operation records. The time-ordered sequence of operation records is encoded using a time-series encoder, transforming each structured operation record into a fixed-length behavior vector, which is then arranged in order to obtain a sequence of asset operation behavior vectors.

[0075] In practical implementation, the transfer time and subject information of multiple nodes in the transfer record set are analyzed, and the original transfer event sequence is constructed in chronological order. Each record in the transfer record set includes the transferring subject, the transferring subject, the asset identifier, and a timestamp. For example, the original transfer event sequence of a certain asset is "[Time T1: Department A → Employee B], [Time T2: Employee B → Project Team C], [Time T3: Project Team C → External Partner D]". Frequently occurring transfer patterns are identified on the original transfer event sequence. The transfer pattern is defined by multiple consecutive transfer events and the subject relationships between the events. For example, if "Department A → Employee B → Project Team C" appears multiple times in the transfer record set, then this consecutive transfer sequence is identified as a transfer pattern and assigned a pattern identifier P001. Based on the identified circulation patterns, the original circulation event sequence is abstracted and compressed to form a high-level circulation logic graph with circulation patterns as the basic unit. For example, three consecutive events in the original sequence can be replaced by a single high-level event "[Time interval T1-T3: Circulation pattern P001]". The nodes of the high-level circulation logic graph are the sets of subjects participating in the pattern, and the edges are the patterned circulation relationships. The high-level circulation logic graph is aligned and associated with the low-level circulation detail graph based on a single transfer event to jointly form a multi-level asset circulation map. For example, the low-level circulation detail graph retains detailed information for each atomic transfer event, while the high-level circulation logic graph stores abstract information at the pattern level. The two are associated through pattern identifiers and event indexes. In some embodiments, the identification of frequent circulation patterns adopts a support counting method based on a time window. For example, a sliding window with a length of K events is set, and the frequency of occurrence of consecutive event sequences in the window is counted. If the frequency exceeds a threshold, it is determined to be a frequent pattern. The definition of the flow pattern allows the inclusion of wildcard subjects. For example, "Department* → Employee X → Project Team Y" can match any sequence from department to employee X and then to project team Y, thus adapting to similar flow behaviors under the same organizational structure.

[0076] In practice, for each access, modification, copy, and deletion operation in the operation log set, a pre-trained natural language understanding model is used to classify the intent and obtain operation intent labels. For example, the operation description "User downloaded file V" is classified as "Read", and the operation description "Administrator updated document W" is classified as "Modify". For each operation, the operation subject, operation time, operation object identifier, and operation parameters are extracted. The operation intent labels and the extracted information are encoded together into a structured operation record. For example, the structured operation record is stored in JSON format, containing fields {"intent":"read","subject":"user123","time":"2025-03-1510:30","object":"file_V","params":{}}. The structured operation records are sorted by operation time to form a time-ordered operation record sequence. For example, operation records for the same asset are arranged in ascending order of timestamps, and the time interval between adjacent records reflects the operation activity. The temporal encoder encodes the time-ordered sequence of operation records, transforming each structured operation record into a fixed-length behavior vector. After arranging them in order, the asset operation behavior vector sequence is obtained. For example, the temporal encoder uses the Transformer architecture to project the discrete features of each operation record into a 128-dimensional vector while retaining the temporal order information.

[0077] It is understandable that pre-trained natural language understanding models are fine-tuned based on textual descriptions in operation logs. For example, training on a labeled operation intent dataset using the BERT architecture can accurately identify the semantics of "access," "modify," "copy," "delete," and their variations. The encoding of operation parameters covers specific operation attributes, such as the source and target paths of copy operations and the list of changed fields for modification operations, all of which are incorporated into the parameter fields of the structured operation record. In some embodiments, the training of the temporal encoder employs a self-supervised prediction task, such as randomly masking some fields of the operation record, requiring the model to reconstruct the masked content based on the context, thereby learning the temporal dependencies of operation behaviors. The dimension of the asset operation behavior vector sequence is consistent with that of the asset feature embedding vector, for example, both being 256-dimensional, facilitating feature alignment in subsequent dynamic fusion networks. Optionally, the support threshold for flow patterns is adjusted according to the business scenario. For example, a lower threshold is set for internal collaboration scenarios to capture common workflows, while a higher threshold is set for cross-organizational flow scenarios to avoid noise interference. The compression ratio of the high-level flow logic graph can be dynamically configured, for example, only patterns that occur more than N times are abstracted, while other events are retained in the lower-level flow detail graph.

[0078] In one embodiment of the present invention, the dynamic fusion network includes a graph encoding branch, a vector alignment branch, and a dynamic memory module. The graph encoding branch receives the asset flow graph and encodes it into graph structure features. The vector alignment branch receives asset feature embedding vectors and asset operation behavior vector sequences. An attention mechanism aligns and fuses the asset feature embedding vectors and asset operation behavior vector sequences along the feature dimension, generating an aligned behavior feature sequence. The dynamic memory module receives the graph structure features and the aligned behavior feature sequence, simulating the evolution of asset states along the time dimension. At each time step, the dynamic memory module updates its internal memory state based on the current behavior features and is constrained by the graph structure features. After the complete time period corresponding to the asset operation behavior vector sequence ends, the final state of the dynamic memory module is read and concatenated with the graph structure features to output a comprehensive state tensor. The asset valuation network includes multiple parallel attention heads, each corresponding to a preset life stage. The comprehensive state tensor is input into the asset valuation network, and each attention head independently weights and focuses on the features in the comprehensive state tensor related to the corresponding life stage. Each attention head outputs a focused feature representation, which is fed into a fully connected layer. The fully connected layer maps the features to a scalar representing the value density at a predetermined lifecycle stage corresponding to the attention head. The scalars output from all the fully connected layers corresponding to all attention heads are aggregated to form the value density distribution of the digital asset across multiple predetermined lifecycle stages.

[0079] In its implementation, the dynamic fusion network comprises a graph encoding branch, a vector alignment branch, and a dynamic memory module. The graph encoding branch receives the asset transfer graph and encodes it into graph structure features. For example, it uses a graph convolutional network to perform hierarchical aggregation of node and edge features in a multi-level asset transfer graph, outputting a fixed-dimensional graph structure feature vector. The vector alignment branch receives asset feature embedding vectors and asset operation behavior vector sequences. Through an attention mechanism, it aligns and fuses the asset feature embedding vectors and asset operation behavior vector sequences along the feature dimension, generating an aligned behavior feature sequence. For example, it uses the asset feature embedding vector as the query vector and the asset operation behavior vector sequence as the key and value vectors. After calculating attention weights, it performs a weighted summation of the behavior vector sequences, aligning the behavior features semantically with the static features of the assets. The dynamic memory module receives graph structure features and aligned behavioral feature sequences, simulating the evolution of asset states over time. At each time step, the dynamic memory module updates its internal memory state based on the current behavioral features and is constrained by the graph structure features. For example, the graph structure features can be used as the initial hidden state of the dynamic memory module. At each time step, the aligned behavioral features of the current time point are input, and the memory state is updated through a gated loop unit, ensuring that the state evolution is always related to the asset flow topology. After the complete time period corresponding to the asset operation behavior vector sequence ends, the final state of the dynamic memory module is read and concatenated with the graph structure features to output a comprehensive state tensor. For example, the final memory state is a 128-dimensional vector, the graph structure features are 128-dimensional vectors, and the concatenation results in a 256-dimensional comprehensive state tensor, covering both the static structure and dynamic behavioral evolution information of the asset.

[0080] In some embodiments, the graph convolutional network of the graph encoding branch is set to 3 layers, aggregating one-hop, two-hop, and three-hop neighbor information layer by layer to ensure that the graph structure features can capture both local connection details and global topological characteristics. The hidden layer dimension of the gated recurrent unit of the dynamic memory module is consistent with the aligned behavioral feature dimension, for example, both are 128-dimensional, to avoid information loss during feature projection.

[0081] In its implementation, the asset valuation network comprises multiple parallel attention heads, each corresponding to a predefined lifecycle stage. These stages include creation, circulation, use, archiving, and destruction. Each attention head focuses on extracting value-related features for its corresponding stage from the comprehensive state tensor. The comprehensive state tensor is input into the asset valuation network, and each attention head independently weights and focuses on the features related to its corresponding lifecycle stage. For example, an attention head targeting the circulation stage increases the weights of features related to circulation paths and transfer frequencies, while suppressing the influence of other irrelevant features. Each attention head outputs a focused feature representation, which is fed into a fully connected layer. This layer maps the features to a scalar representing the value density at the predefined lifecycle stage corresponding to the attention head. For instance, the fully connected layer uses a linear transformation and a sigmoid activation function to map the 128-dimensional focused features to scalar values ​​between 0 and 1, with higher values ​​indicating greater value density at that stage. The scalars of the fully connected layer outputs corresponding to all attention heads are aggregated to form the value density distribution of digital assets in multiple preset life stages. For example, the output scalars of the 5 attention heads are arranged into a vector [0.85, 0.92, 0.78, 0.35, 0.12] in the order of the stages, which correspond to the value density of the creation, circulation, use, archiving and destruction stages, respectively.

[0082] It is understandable that the number of attention heads is exactly the same as the number of preset lifecycle stages, with each stage assigned a unique attention head to ensure that each dimension of the value density distribution has a clear physical meaning. The parameters of the fully connected layers are learned independently, adapting to the differences in value assessment standards across different lifecycle stages. For example, the value assessment in the usage stage relies more on operational behavior characteristics, while the archiving stage relies more on asset type and permission characteristics. In some embodiments, the weighted focusing process of the attention heads employs a scaled dot product attention mechanism, introducing a stage-specific parameter matrix when calculating attention weights, allowing different attention heads to focus on different feature subspaces of the comprehensive state tensor. The scalar aggregation of the value density distribution uses a concatenation operation to form a density vector corresponding one-to-one with the lifecycle stage, facilitating direct use by the subsequent policy generation model.

[0083] Optionally, the predefined lifecycle stages can be expanded or merged according to business needs. For example, a "Pending Approval" stage can be added, or "Archived" and "Destroyed" can be merged into a "Final Stage." This only requires adjusting the number of attention heads and the output dimension of the fully connected layer in the asset valuation network. The dimensional differences between the comprehensive state tensor and the value density distribution are adapted through a projection layer. For example, if the comprehensive state tensor is 256-dimensional and the query vector dimension of each attention head is set to 64-dimensional, dimensional alignment is achieved through linear transformation, as shown in Table 1.

[0084] Table 1: Network Attention Focus Allocation Table for Asset Valuation

[0085] Life stages Attention head icon Query vector dimension Key vector dimension Value vector dimension Creation phase Head_Create 64 64 64 circulation stage Head_Transfer 64 64 64 Use phase Head_Use 64 64 64 Archiving stage Head_Archive 64 64 64 Destruction phase Head_Destroy 64 64 64

[0086] In practical implementation, the weighted focusing process for each attention head can be calculated using the following formula:

[0087]

[0088] in: Indicates the first The focused feature representation of the output of each attention head. It is the first The query matrix corresponding to each attention head is obtained by performing a stage-specific linear transformation on the comprehensive state tensor. It is the key matrix shared by all attention heads, obtained by linear transformation of the integrated state tensor. It is a value matrix shared by all attention heads, obtained by linear transformation of the integrated state tensor. is the dimension of the key vector, used to scale the dot product result to prevent gradient vanishing. This formula ensures that each attention head filters relevant features from the comprehensive state tensor based on the stage-specific query, forming a stage-specific feature-focused representation.

[0089] In one embodiment of the present invention, the value density distribution, the comprehensive state tensor, and the preset management objective constraints are input into the strategy generation model. Under the management objective constraints and guided by the value density distribution, the strategy generation model searches for the optimal management action for each lifecycle stage in the entire strategy space from creation to destruction. The strategy generation model assembles the optimal management action found in each lifecycle stage into a complete management strategy path by combining the transition logic between lifecycle stages. The management strategy path is evaluated and post-processed, transforming it into an adaptive management strategy set composed of specific management rules, triggering conditions, and execution actions. Each lifecycle stage is considered a decision node, and a full-process decision graph from creation to destruction is constructed. In the full-process decision graph, the transition cost between nodes is defined according to the management objective constraints, and the state value of the node itself is defined according to the value density distribution. An improved graph search algorithm is run on the full-process decision graph. During the search process, the improved graph search algorithm simultaneously considers the state value of the current node, the cost of transitioning to the next node, and the estimated value from the next node to the endpoint. The search path output by the improved graph search algorithm has transition edges selected at each decision node representing the optimal management action found in the lifecycle stage.

[0090] In practical implementation, the value density distribution, the comprehensive state tensor, and preset management objective constraints are input into the strategy generation model. These preset constraints include boundary conditions such as compliance requirements, storage cost caps, and access efficiency lower limits. For example, compliance requirements stipulate that sensitive data must be destroyed within three years of archiving, and the storage cost cap limits the total storage cost of a single asset throughout its lifecycle to a specified threshold. Under these management objective constraints, the strategy generation model, guided by the value density distribution, searches for the optimal management action for each lifecycle stage within the entire strategy space from creation to destruction. For instance, if the value density distribution shows that the value density of the usage stage is significantly higher than that of the archiving stage, the validity period of the usage stage is extended instead of immediate archiving. The strategy generation model then combines the optimal management actions found for each lifecycle stage with the transition logic between lifecycle stages to assemble a complete management strategy path. For example, a complete management strategy path includes a sequence of actions such as "enabling version control during creation," "mandating watermarking during circulation," "automatic backup during usage," "encryption and compression during archiving," and "physical erasure during destruction." The management strategy path is evaluated and post-processed, and transformed into an adaptive management strategy set consisting of specific management rules, triggering conditions and execution actions. For example, the management rule is "If the asset is in the circulation stage and the value density is less than 0.3, it will enter the archiving review in advance", the triggering condition is "No access operation for 30 consecutive days", and the execution action is "Start the archiving workflow".

[0091] In some embodiments, management objective constraints are presented in the form of linear inequalities. For example, the total storage cost constraint is expressed as the sum of the products of storage duration and unit time cost at each stage being less than or equal to the budget limit, and the access efficiency constraint requires the average response time to be less than a specified number of milliseconds. The search space of the policy generation model covers all legal lifecycle stage transition paths. For example, a direct transition from the creation stage to the circulation or usage stage is possible, but a transition from the archiving stage can only be to the destruction stage or retain the archived state.

[0092] In practical implementation, each lifecycle stage is treated as a decision node, constructing a full-process decision graph from creation to destruction. The nodes in this graph include creation, transfer, use, archiving, and destruction nodes, with edges representing legitimate transfer paths between stages. Within the full-process decision graph, the transfer cost between nodes is defined based on management objective constraints, and the node's own state value is defined based on value density distribution. For example, the transfer cost from a use node to an archive node includes data migration costs and compliance verification time. The state value of a use node is calculated by multiplying the use stage value density in the value density distribution by the expected remaining usage time. An improved graph search algorithm is run on the full-process decision graph. During the search process, this algorithm simultaneously considers the current node's state value, the cost of transferring to the next node, and the estimated value from the next node to the destination. For example, the estimated value is estimated using a heuristic function, comprehensively considering the potential value density of the next node and the minimum cost of subsequent paths. The improved graph search algorithm outputs a search path where the selected transition edge at each decision node represents the optimal management action found in the lifecycle stage. For example, the optimal management action corresponding to the edge from the user node to the archive node in the search path is "start the archive process when the triggering condition is met".

[0093] It is understood that the number of nodes in the entire process decision graph is consistent with the number of preset life stages, and the edge connections follow the natural order of the life cycle, prohibiting reverse transfers or illegal jumps. The calculation of node state value integrates value density distribution and stage characteristics. For example, the state value of the creation stage mainly depends on the expected recovery of initial investment costs, while the circulation stage depends on the diffusion value brought by the number of circulations and the scope. In some embodiments, the improved graph search algorithm adopts a variant of the A* algorithm, using node state value as a component of a heuristic function to guide the search towards maximizing overall value. The calculation of transfer costs supports dynamic adjustment. For example, when the unit price of storage hardware decreases, the storage cost coefficient of the archiving stage is reduced, making the strategy more inclined to extend the archiving time. Optionally, the post-processing of the management strategy path includes redundant action merging and conflict detection. For example, the "encryption" actions of multiple consecutive stages are merged into a single global encryption configuration to avoid repeated operations that increase system overhead. The output format of the adaptive management strategy set adopts a structured configuration file, which includes rule ID, trigger expression, action list, and priority field, making it easy for downstream systems to parse and execute. See Table 2.

[0094] Table 2: Value and Transfer Cost of Nodes in the Whole-Process Decision-Making Diagram

[0095] Decision Node Basis for calculating state value Typical transfer edge Components of transfer costs Create a node Initial investment cost × recovery factor Creation → Flow Verification cost + permission configuration time Flow node Number of circulations × Value gain per transaction Circulation → Use Watermark injection cost + audit log overhead Use nodes Value density × Expected duration Use → Archive Data compression time + encryption computation cost Archive nodes Compliance Preservation Value - Storage Cost Archive → Destroy Secure erasure time + compliant deregistration fees Destroy Nodes Fixed termination value none none

[0096] In practical implementation, the value evaluation function of the improved graph search algorithm adopts the following formula:

[0097]

[0098] in: This indicates the nodes passed from the starting point. The estimated total value upon reaching the destination. From the starting point to the node The total actual cumulative transfer cost From node The estimated minimum transfer cost to the destination. It is a node The value of its own state. This formula balances the costs already incurred, the estimated costs of the future, and the value of the current node during the search process, ensuring that the final search path optimizes the overall value in the entire policy space.

[0099] In one embodiment of the present invention, an adaptive management strategy set is transformed into an executable atomic operation instruction stream and pushed to the target digital asset management platform to drive it to perform corresponding full lifecycle management operations on the target digital assets. Each management rule, triggering condition, and execution action in the adaptive management strategy set is parsed. Each management rule is decomposed into a logical judgment unit, each triggering condition is mapped to an event listener configuration, and each execution action is translated into a call instruction for the underlying interface of the target digital asset management platform. Based on the logical execution order of the management strategy, the logical judgment units, event listener configurations, and underlying interface call instructions are arranged to generate an instruction sequence with clear execution logic and dependencies. The instruction sequence is optimized, and operations that can be executed in parallel are merged to form the final atomic operation instruction stream.

[0100] In practical implementation, the adaptive management strategy set is transformed into an executable atomic operation instruction stream and pushed to the target digital asset management platform. This drives the target digital asset management platform to perform corresponding full lifecycle management operations on the target digital assets. For example, if the target digital asset management platform is an enterprise-level content management system with a RESTful API interface, it can receive instructions and schedule the underlying execution engine to complete asset management tasks. Each management rule, triggering condition, and execution action in the adaptive management strategy set is parsed. For example, one management rule in the adaptive management strategy set is "If the asset is in use and has been accessed less than 5 times in the last 30 days, then start the archiving process." The triggering condition is "Time condition: Scan at 2:00 AM daily; Event condition: Access log update." The execution action is "Call the archiving service interface, parameters: Asset ID=Asset_001, Compression level=High, Encryption algorithm=AES-256." Each management rule is broken down into logical judgment units, each triggering condition is mapped to an event listener configuration, and each execution action is translated into a call instruction for the underlying interface of the target digital asset management platform. For example, the logical judgment unit is converted into the SQL query condition "SELECT COUNT(*) FROM access_log WHERE asset_id='Asset_001' AND time>NOW()-INTERVAL30DAY", the event listener configuration is converted into the message queue subscription parameter "topic:audit.log.update,filter:asset_id=Asset_001", and the underlying interface call instruction is converted into the HTTP request "POST / api / v1 / assets / archive,body:{'assetId':'Asset_001','options':{'compression':'high','encryption':'aes256'}}".

[0101] In some embodiments, the implementation of the logical judgment unit relies on the query engine of the target digital asset management platform. For example, the conditional expressions in the rules are compiled into Elasticsearch query DSL or database SQL statements to support real-time matching of asset status changes. Event listeners are configured to integrate with the platform's event bus, such as registering as a Kafka consumer or RabbitMQ listener, triggering rule logic execution when specified events occur. The underlying interface call instructions strictly adhere to the platform's API specifications, including authentication tokens, parameter encoding, and error retry strategies. For example, each call instruction carries an OAuth2.0 BearerToken, parameters are serialized in JSON format, and retry three times using an exponential backoff strategy upon failure.

[0102] In practical implementation, based on the logical execution order of the management strategy, the logical judgment unit, event listener configuration, and underlying interface call instructions are orchestrated to generate an instruction sequence with clear execution logic and dependencies. For example, the structure of the instruction sequence is: [Listener Startup] → [Timed Trigger Condition Check] → [Execution Logic Judgment Unit] → [If Condition is Met, Send Interface Call Instruction] → [Wait for Interface Return] → [Record Execution Log]. The instruction sequence is optimized by merging operations that can be executed in parallel to form the final atomic operation instruction flow. For example, when multiple rules are detected to share the same event listener configuration, they are merged into a single listener instance and the event is distributed to different rule handlers. Alternatively, multiple interface calls to the same batch of assets can be merged into a batch request to reduce network round-trip overhead.

[0103] It is understandable that the instruction sequence is arranged using a directed acyclic graph structure, where nodes represent logical judgment units or interface call instructions, and edges represent execution dependencies. For example, the execution of subsequent nodes is only triggered when the output state of the previous node is true. The optimization process of the atomic operation instruction flow follows the principle of data locality. For example, multiple instructions that need to be executed on the same server are packaged into a script and distributed at once to reduce the number of remote calls. In some embodiments, the optimization of the instruction sequence includes removing redundant condition judgments and merging duplicate resource operations. For example, if two rules both need to check the number of asset accesses, they can be merged into a single query and the results can be reused. The final format of the atomic operation instruction flow adopts a platform-parsable intermediate language, such as a JSON serialized workflow definition or a YAML-configured automated task list, ensuring that the target digital asset management platform can directly schedule and execute it without additional conversion.

[0104] Optionally, the atomic operation instruction stream can be pushed in two modes: real-time streaming and batch synchronization. For example, the instruction stream generated by high-priority strategies can be pushed to the platform in real time via WebSocket, while low-priority strategies can be synchronized through periodic batch file uploads. The target digital asset management platform provides status codes and execution durations when executing the instruction stream, which are used for subsequent reverse optimization of the strategy model. For example, it records the success rate of interface calls and latency metrics, serving as a basis for adjusting the transfer cost of the strategy generation model.

[0105] In practical implementation, the benefits of parallel merging optimization of atomic operation instruction streams can be quantified using the following formula:

[0106]

[0107] in: This represents the total round-trip time saved by merging and optimization. This is the number of serial instructions before optimization. It is the average round-trip time from the issuance of a single command to the response. This represents the upper limit of the instruction capacity for batch processing. This formula measures the theoretical time gain from reducing the number of network interactions through batch merging, and guides the parameter adjustment of optimization strategies.

[0108] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A digital asset lifecycle management method, characterized in that, The method includes: Obtain the metadata set, transfer record set, and operation log set of the target digital asset. The metadata set includes an asset content summary, asset type tags, and a list of permission definitions. The transfer record set includes the transfer time and subject information of multiple nodes. The operation log set includes access, modification, copy, and deletion operation information at multiple times. An improved graph embedding algorithm is applied to the metadata set to generate asset feature embedding vectors. The improved graph embedding algorithm optimizes the traditional embedding algorithm based on the semantic association and circulation topology between digital assets. The set of transfer records is processed by structural pattern extraction to generate a multi-level asset transfer graph. The set of operation logs is processed by intent parsing and behavior encoding to generate an asset operation behavior vector sequence. The asset feature embedding vector, asset flow map, and asset operation behavior vector sequence are input into a dynamic fusion network to generate a comprehensive state tensor representing the dynamics of the entire digital asset lifecycle. The attention-based asset valuation network is invoked to analyze the comprehensive state tensor and output the value density distribution of digital assets in multiple preset life stages. Based on the value density distribution, an adaptive management strategy set covering the entire process of digital assets from creation, circulation, use, archiving to destruction is calculated and output through a strategy generation model.

2. The method for managing the entire lifecycle of digital assets according to claim 1, characterized in that, An improved graph embedding algorithm is performed on the metadata set to generate asset feature embedding vectors, including: Based on the asset type tags and permission definition list in the metadata set, an initial asset relationship graph is constructed with digital assets as nodes and semantic associations and permission inheritance relationships between assets as edges; On the initial asset relationship graph, historical transfer paths between assets are added as reinforcing edges based on the set of transfer records to form an enhanced asset relationship graph; Based on the improved graph embedding algorithm, multi-hop neighbor information aggregation is performed on the enhanced asset relationship graph. During the aggregation process, the improved graph embedding algorithm defines differentiated information transmission weights for different types of edges. The differentiated information transmission weights are jointly determined by the semantic association strength, the permission inheritance depth, and the historical flow frequency. After completing multiple rounds of information aggregation, the feature vector corresponding to each asset node is sampled and normalized, and the asset feature embedding vector corresponding to each digital asset is output.

3. The method for full lifecycle management of digital assets according to claim 1, characterized in that, The structure pattern extraction process is performed on the set of transfer records to generate a multi-level asset transfer map, including: Analyze the multi-node transfer time and main information in the set of transfer records, and construct the original transfer event sequence in chronological order; On the original sequence of events, frequently occurring patterns of transfer are identified. These patterns are defined by multiple consecutive transfer events and the relationships between the events. Based on the identified flow patterns, the original flow event sequence is abstracted and compressed to form a high-level flow logic diagram with flow patterns as the basic unit. Align and associate the high-level transfer logic diagram with the low-level transfer detail diagram based on a single transfer event to jointly form the multi-level asset transfer map.

4. The method for managing the entire lifecycle of digital assets according to claim 1, characterized in that, The operation log set is subjected to intent parsing and behavior encoding processing to generate an asset operation behavior vector sequence, including: For each access, modification, copy, and deletion operation in the operation log set, a pre-trained natural language understanding model is invoked to perform intent classification and obtain the operation intent label; For each operation message, extract its operation subject, operation time, operation object identifier and operation parameters, and encode the operation intention tag and the extracted information together into a structured operation record; The structured operation records are sorted according to the operation time to form a time-ordered sequence of operation records; The time-ordered sequence of operation records is encoded by a time encoder, and each structured operation record is transformed into a fixed-length behavior vector. The vectors are then arranged in order to obtain the asset operation behavior vector sequence.

5. The method for full lifecycle management of digital assets according to claim 1, characterized in that, The asset feature embedding vector, asset circulation map, and asset operation behavior vector sequence are input into a dynamic fusion network to generate a comprehensive state tensor representing the dynamics of the entire digital asset lifecycle, including: The dynamic fusion network includes a graph coding branch, a vector alignment branch, and a dynamic memory module; The graph encoding branch receives the asset transfer graph and encodes it into graph structure features; The vector alignment branch receives the asset feature embedding vector and the asset operation behavior vector sequence; The asset feature embedding vector and the asset operation behavior vector sequence are aligned and fused in the feature dimension by using an attention mechanism to generate an aligned behavior feature sequence. The dynamic memory module receives the graph structure features and the aligned behavioral feature sequence, and simulates the evolution of asset state in the time dimension. At each time step, the dynamic memory module updates its internal memory state based on the current behavioral features and is constrained by the graph structure features. After the complete time period corresponding to the asset operation behavior vector sequence ends, the final state of the dynamic memory module is read and concatenated with the graph structure features to output the comprehensive state tensor.

6. The method for managing the entire lifecycle of digital assets according to claim 1, characterized in that, The invocation of an attention-based asset valuation network to analyze the comprehensive state tensor outputs the value density distribution of the digital asset across multiple preset life stages, including: The asset valuation network contains multiple parallel attention heads, each corresponding to a preset life stage; The comprehensive state tensor is input into the asset valuation network, and each attention head independently weights and focuses on the features in the comprehensive state tensor that are related to the corresponding life stage. Each attention head outputs a focused feature representation, which is fed into a fully connected layer that maps the features to a scalar representing the value density of the preset life stage corresponding to the attention head. The scalars output from the fully connected layer corresponding to all attention points are aggregated to form the value density distribution of the digital asset across multiple preset life stages.

7. The method for full lifecycle management of digital assets according to claim 1, characterized in that, Based on the value density distribution, a set of adaptive management strategies covering the entire process of digital assets from creation, circulation, use, archiving to destruction is calculated and output through a strategy generation model, including: The value density distribution, the comprehensive state tensor, and the preset management objective constraints are input into the strategy generation model; Under the constraints of the management objectives and guided by the value density distribution, the strategy generation model searches for the optimal management action for each life stage in the entire strategy space from creation to destruction. The strategy generation model assembles the optimal management action found in each life stage into a complete management strategy path by combining the transition logic between life stages. The management strategy path is evaluated and post-processed to transform it into an adaptive management strategy set consisting of specific management rules, triggering conditions, and execution actions.

8. The method for full lifecycle management of digital assets according to claim 7, characterized in that, Under the constraints of the management objectives and guided by the value density distribution, the strategy generation model searches for the optimal management action for each lifecycle stage in the entire strategy space from creation to destruction, including: Treat each life stage as a decision node and construct a decision graph for the entire process from creation to destruction; In the full-process decision diagram, the transfer cost between nodes is defined based on the management objective constraints, and the state value of each node is defined based on the value density distribution. An improved graph search algorithm is run on the entire process decision graph. During the search process, the improved graph search algorithm simultaneously considers the state value of the current node, the cost of transitioning to the next node, and the estimated value from the next node to the destination. The improved graph search algorithm outputs a search path, where the selected transition edge at each decision node represents the optimal management action found in the life stage.

9. The method for full lifecycle management of digital assets according to claim 1, characterized in that, The method further includes: The adaptive management strategy set is transformed into an executable atomic operation instruction stream and pushed to the target digital asset management platform to drive it to perform corresponding full lifecycle management operations on the target digital asset. The adaptive management strategy set is transformed into an executable stream of atomic operation instructions, including: Analyze each management rule, triggering condition, and execution action in the set of adaptive management strategies; Each management rule is broken down into logical judgment units, each triggering condition is mapped to an event listener configuration, and each execution action is translated into a call instruction for the underlying interface of the target digital asset management platform. Based on the logical execution order of the management strategy, the logical judgment unit, event listener configuration, and underlying interface call instructions are arranged to generate an instruction sequence with clear execution logic and dependencies; The instruction sequence is optimized by merging operations that can be executed in parallel to form the final atomic operation instruction stream.

10. A digital asset lifecycle management system, characterized in that, The system includes a processor and a memory, the memory and the processor being connected. The memory is used to store programs, instructions or code, and the processor is used to run the programs, instructions or code in the memory to implement the digital asset lifecycle management method according to any one of claims 1-9.

Images