A method and system for managing digital assets based on the metaverse

Abstract

The application relates to the technical field of digital asset management, in particular to a digital asset management method and system based on a meta universe, which comprises the following steps: constructing a basic value curve and a normal market fluctuation range of different types of assets based on historical digital asset interaction records; performing value deviation detection on real-time transaction data based on the dynamic benchmark to identify value abnormal assets; tracking user behavior trajectories of the abnormal assets to generate an asset attention curve and a user behavior coordination mode matrix; inputting the above data into an asset state analysis model based on a graph neural network to obtain a multi-dimensional state abnormal probability distribution graph; and determining a final management strategy level according to the aggregation form and intensity of the probability distribution. The method can accurately identify hidden market manipulation and systematic risk through dynamic value benchmarking and collaborative behavior analysis driven by a graph model, and the intelligentization and accuracy of meta universe digital asset management are improved.

Description

Technical Field

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

[0002] Currently, the mainstream methods for managing metaverse digital assets typically rely on simple transaction price thresholds or static historical average value models for anomaly monitoring. These systems trigger alerts by setting fixed price fluctuation ranges or depending on a single-dimensional surge in trading volume. Another common approach is to analyze individual user behavior logs, such as login frequency or single transaction amount, to identify potentially malicious accounts. These technologies form the foundation of current metaverse asset risk prevention and control.

[0003] Existing solutions have shortcomings. Static threshold models cannot adapt to the dynamic evolution of digital asset value in the metaverse, which is influenced by spatial events, content updates, or community sentiment. They are prone to misjudging normal market fluctuations as anomalies or failing to detect manipulation behaviors that occur within a slow trend. Analyzing only individual user behavior completely ignores the complex patterns of asset value manipulation within the metaverse, where users manipulate through community collaboration and multi-point synchronous operations. Such collaborative behaviors may appear normal on an individual level, but their aggregation can have a significant impact, and existing methods lack the ability to effectively detect this.

[0004] Establishing a monitoring system capable of understanding dynamic benchmarks for asset value and penetrating the surface of individual behavior to uncover hidden collaborative relationships among users has become a key challenge for accurately identifying the true state of digital assets in the metaverse and implementing effective hierarchical management. This requires the system to not only assess whether the fluctuations of the asset itself are abnormal, but also to diagnose the structural anomalies in the behavior of user groups behind these fluctuations. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a digital asset management method and system based on the metaverse.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a digital asset management method based on the metaverse, comprising:

[0007] Historical digital asset interaction records are extracted from multi-source metaverse space data, and asset type classification and basic value modeling are performed on them to generate basic value curves and normal market fluctuation range parameter sets corresponding to different asset types.

[0008] Receive real-time data streams of target digital asset transactions and usage within the metaverse, and perform dynamic value deviation detection on the target digital asset transaction and usage data streams based on the basic value curve and the normal market fluctuation range parameter set. Mark digital assets whose value fluctuations exceed the corresponding normal market fluctuation range parameters to obtain assets with abnormal value.

[0009] The user behavior trajectory of the aforementioned abnormal value assets is tracked and aggregated within a continuous time window to generate an asset attention curve and a user behavior collaborative pattern matrix that evolves over time.

[0010] The asset attention curve and user behavior collaborative pattern matrix are input into the asset status analysis model based on graph neural network. The asset status analysis model outputs the probability distribution map of the target digital asset's status anomaly across multiple evaluation dimensions.

[0011] Based on the clustering pattern and intensity of probability values ​​in the aforementioned abnormal state probability distribution diagram, the final management strategy level of the target digital asset is determined.

[0012] As a further aspect of the present invention, the asset type classification and basic value modeling include:

[0013] Digital assets in the metaverse are divided into multiple asset classes with different economic attributes using predefined metadata standards;

[0014] For each asset class, transaction and usage data of the corresponding asset class under normal market conditions are extracted from the historical digital asset interaction records;

[0015] Value characteristic analysis and volatility statistics are performed on the transaction and usage data of the corresponding asset classes under normal market conditions to establish a basic value trend line and market volatility statistical distribution chart for each asset class;

[0016] Based on the aforementioned fundamental value trend line and market volatility statistical distribution chart, the normal market volatility range parameter for each asset class is defined.

[0017] As a further aspect of the present invention, the dynamic value deviation detection of the target digital asset transaction and usage data stream based on the basic value curve and the normal market fluctuation range parameter set includes:

[0018] The real-time collected data stream of target digital asset transactions and usage within the metaverse is divided into continuous data segments in chronological order.

[0019] For each data segment, the difference between its value index and the basic value trend line is calculated under the corresponding asset class to obtain the absolute value deviation.

[0020] Simultaneously, the rate of change of value of the current data segment compared to the previous data segment within the corresponding asset class is calculated to obtain the value volatility.

[0021] The absolute deviation of value and the volatility of value are compared with the normal market volatility range parameters defined in the market volatility statistical distribution chart, respectively.

[0022] If either the absolute deviation of value or the volatility of value continuously exceeds the corresponding normal market fluctuation range parameter for a preset duration, the current asset will be identified as an asset with abnormal value.

[0023] As a further aspect of the present invention, the step of tracking and aggregating user behavior trajectories within a continuous time window for the abnormally valuable assets includes:

[0024] For each identified anomalous asset, starting from the time of its identification, user interaction behavior features surrounding the anomalous asset are extracted, including access frequency, holder change chain, and intensity of associated social topics, to form an initial behavioral feature vector.

[0025] In subsequent consecutive time windows, even if the value indicators of the abnormal assets return to within the normal market fluctuation range parameters, the same behavioral characteristics will continue to be tracked and extracted to form a sequence of behavioral characteristic vectors.

[0026] For each feature dimension in the behavioral feature vector sequence, calculate the trajectory of its value change over time to generate multiple behavioral feature change trajectory lines;

[0027] By smoothing and intensity accumulation calculations on all behavioral characteristic change trajectories, the asset attention curve representing the sustained popularity of the behavior is obtained.

[0028] As a further aspect of the present invention, the generation of the asset attention curve and user behavior collaborative pattern matrix that evolve over time includes:

[0029] From the sequence of behavioral feature vectors, analyze the synchronicity coefficient of the changing trends of any two different behavioral feature dimensions at different points in time;

[0030] Arrange the pairwise synchronization coefficients of all behavioral feature dimensions in the order of the feature dimensions to construct the initial behavioral synchronization matrix;

[0031] The initial behavior synchronization matrix is ​​subjected to a time-dimensional moving average, and the mean of the synchronization coefficient within each moving window is calculated to form a dynamic behavior synchronization matrix sequence.

[0032] The corresponding elements of all matrices in the dynamic behavior coordination matrix sequence are merged to finally generate the user behavior coordination pattern matrix that reflects the complex correlation pattern of user behavior.

[0033] As a further aspect of the present invention, the construction and operation of the asset status analysis model based on graph neural networks includes:

[0034] A model input interface is constructed to receive the asset attention curve and the user behavior collaboration pattern matrix, and the user behavior collaboration pattern matrix is ​​used as the initial weight of the edges between nodes in the graph.

[0035] Construct an asset-behavior heterogeneous graph, with digital assets and user behavior features as two types of nodes, and the association between assets and behaviors and the synergistic relationship between behaviors as edges;

[0036] A graph attention network layer is constructed to perform multi-round information aggregation and updating on nodes and edges in the asset-behavior heterogeneous graph, and to learn the high-order embedding representation of nodes.

[0037] A graph structure readout and decision layer is constructed, the updated asset node embedded representation is pooled, and mapped to the state anomaly probability space through a multilayer perceptron, outputting the state anomaly probability distribution map.

[0038] As a further aspect of the present invention, the workflow of the graph attention network layer includes:

[0039] For each target node in the asset-behavior heterogeneous graph, calculate the vector representation of the features of all its neighboring nodes after linear transformation;

[0040] Calculate the attention coefficient between the target node and the feature vector of each neighbor node. The attention coefficient is determined by the node features and the weights of the connecting edges.

[0041] The calculated attention coefficients are processed by a normalization function to obtain standardized attention weights;

[0042] The standardized attention weights are used to perform a weighted summation of the feature vectors of neighboring nodes, combined with the target node's own features, and then processed through a nonlinear activation function to generate a new feature representation of the target node.

[0043] As a further aspect of the present invention, the determination of the final management strategy level of the target digital asset includes:

[0044] Analyze the state anomaly probability distribution map, identify continuous regions in the map where the probability value exceeds a set threshold, and define the continuous regions as high anomaly risk clusters;

[0045] Calculate the core metrics for each high-risk cluster, including the duration of the abnormal state, the number of associated user behavior feature dimensions, and the average anomaly probability intensity within the cluster.

[0046] A management strategy level mapping rule base is established, in which management strategy levels are defined for different combinations of core indicators;

[0047] The core indicators of each high-risk cluster are matched with the management strategy level mapping rule base, and a candidate management strategy level is assigned to each cluster;

[0048] Among all the candidate management strategy levels assigned to high-risk clusters, the final management strategy level applied to the digital asset is determined according to the preset conflict resolution rules.

[0049] As a further aspect of the present invention, the step of performing a time-dimensional moving average processing on the initial behavior synchronization matrix to calculate the mean of the synchronization coefficients within each moving window, forming a dynamic behavior synchronization matrix sequence, includes:

[0050] A fixed-length time sliding window is set, and the initial behavior synchronization matrix is ​​divided into multiple time segments according to the timestamp order;

[0051] All initial behavior synchronization matrices within each time segment are aligned and superimposed according to the matrix element positions, and the arithmetic mean of all synchronization coefficients at each matrix position is calculated.

[0052] The calculated arithmetic mean is subjected to matrix reconstruction to generate the average behavior synchronization matrix corresponding to the current time sliding window.

[0053] The average behavior synchronization matrices corresponding to each time sliding window are arranged and combined in chronological order to form a dynamic behavior synchronization matrix sequence that evolves over time.

[0054] As a further aspect of the present invention, the present invention also includes a digital asset management system based on a metaverse, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, it implements the steps of the digital asset management method based on a metaverse as described above.

[0055] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0056] By tracking and aggregating user behavior trajectories of assets with abnormal value within a continuous time window, a user behavior collaborative pattern matrix that evolves over time is generated. This matrix, along with the asset attention curve, is input into an asset state analysis model based on a graph neural network. The graph neural network can model the complex relationships and propagation paths between user behaviors, thereby resolving dispersed but collaborative group behavior patterns. The model outputs a multi-dimensional state anomaly probability distribution map, which probabilistically characterizes the probability of anomalies and their correlation strength across different dimensions such as liquidity, manipulation risk, and consensus stability. This achieves a shift from detecting single anomalies to assessing systemic risk, enabling managers to identify different anomaly patterns caused by random market noise, healthy market hype, and organized manipulation based on the shape and clustering strength of the probability distribution.

[0057] Based on a dynamic underlying value curve and a set of parameters for normal market fluctuation ranges, learned from multi-source historical data and matched to the asset type, the system detects value deviations in real-time data streams. This mechanism means that the system's criteria for judging "anomalies" are no longer fixed values, but rather dynamic benchmarks that adaptively adjust according to the asset type, its life cycle stage, and historical normal fluctuation patterns. It can effectively filter out short-term reasonable value fluctuations caused by specific activities within the metaverse, while keenly capturing hidden value drifts or manipulation signals that deviate from their own historical normal behavior patterns. This improves the targeting and accuracy of anomaly detection, providing a reliable basis for the subsequent precise implementation of differentiated management strategies. Attached Figure Description

[0058] Figure 1 This is a flowchart of the digital asset management method based on the metaverse described in this invention;

[0059] Figure 2 A flowchart for classifying asset types and modeling underlying value;

[0060] Figure 3 A flowchart for user behavior tracking and aggregation;

[0061] Figure 4 Heatmap of user behavior collaboration patterns;

[0062] Figure 5 This is a trend curve showing the change in attention given to digital assets in the metaverse over time. 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 process extracts historical digital asset interaction records from multi-source metaverse spatial data, classifies these records into asset types, and models their fundamental value, generating fundamental value curves and normal market fluctuation range parameter sets corresponding to different asset types. It receives real-time data streams of target digital asset transactions and usage within the metaverse, and dynamically detects value deviations based on the fundamental value curves and normal market fluctuation range parameter sets. Digital assets whose value fluctuations exceed the corresponding normal market fluctuation range parameters are identified as having abnormal value. User behavior trajectories of these abnormal value assets are tracked and aggregated within a continuous time window, generating an asset attention curve and a user behavior collaboration pattern matrix that evolves over time. These curves are then input into a graph neural network-based asset state analysis model, which outputs a probability distribution map of the target digital asset's abnormal state across multiple evaluation dimensions. Based on the clustering pattern and intensity of probability values ​​in the abnormal state probability distribution map, the final management strategy level for the target digital asset is determined.

[0066] See Figure 2In one embodiment of the present invention, during the operation of the metaverse platform, multi-source metaverse space data, including virtual mall transaction logs, user asset usage records, and cross-scenario interactive events, is aggregated. Historical digital asset interaction records are extracted from these data. In specific implementations, a predefined metadata standard is used to classify digital assets in the metaverse into multiple asset categories with different economic attributes. These economic attributes cover different categories such as tradable virtual items, functional virtual props, identity-based assets, and composite rights certificates. For each asset category, transaction and usage data under normal market conditions are extracted from historical digital asset interaction records. Normal market conditions refer to a period without sudden intervention and where the number of participating users and their activity levels are at the median level of the platform's statistics. In practice, value characteristic analysis and volatility statistics are performed on the transaction and usage data of corresponding asset classes under normal market conditions. Value characteristic analysis includes calculating the average transaction price per unit time, the distribution of transaction frequency, and the proportion of usage time. Volatility statistics include measuring the standard deviation and quantile intervals. Based on the analysis results, a basic value trend line and a market volatility statistical distribution map are established for each asset class. The basic value trend line shows the smooth trend of the corresponding asset class's value over time under normal conditions, and the market volatility statistical distribution map records the probability of occurrence of fluctuations of different magnitudes and their duration. Combining the basic value trend line and the market volatility statistical distribution map, normal market volatility range parameters are defined for each asset class. The normal market volatility range parameters include two values: the upper limit of absolute deviation and the upper limit of volatility, which correspond to the acceptable deviation of the value benchmark and the acceptable threshold of the rate of value change in adjacent periods, respectively.

[0067] In practice, the system receives real-time data streams of target digital asset transactions and usage within the metaverse. These data streams are updated at fixed collection cycles and include asset identifiers, transaction prices, usage frequency, and scenario distribution information. The real-time data streams are then segmented chronologically into continuous data segments, each covering an equal-length observation interval. For each data segment, the difference between its value indicators and the basic value trend line is calculated under the corresponding asset category to obtain the absolute value deviation. The absolute value deviation is calculated as the absolute value of the difference between the arithmetic mean of the value indicators within the data segment and the theoretical value of the basic value trend line at the same time point, which can be expressed by the following formula:

[0068]

[0069] in: Indicates the absolute deviation of value. This indicates the number of sampling points within the data segment. Indicates the first The value index value of each sampling point This indicates the fundamental value trend line at the center of the segment. The theoretical value is calculated. Simultaneously, the rate of change of value between the current data segment and the previous data segment within the corresponding asset class is calculated to obtain the value volatility, which is the relative increment of the mean value of the current segment compared to the mean value of the previous segment. The absolute value deviation and value volatility are then compared with the normal market volatility range parameters defined in the market volatility statistical distribution chart. The normal market volatility range parameters have previously been set with independent upper limits for absolute deviation and volatility for each asset class.

[0070] In practical implementation, if either the absolute value deviation or the value volatility consistently exceeds the corresponding normal market volatility range parameter for a preset duration, the current asset is identified as an anomalous asset. The preset duration is set based on the typical duration of anomalous fluctuations in historical data for the asset class. In some embodiments, the upper limit of absolute deviation and the upper limit of volatility can be adjusted by the platform operator according to regulatory requirements and business stability needs, making the detection process conform to the actual operating environment. In some embodiments, the segment length of the data can be selected based on the trading activity cycle of the asset class; shorter segments are used for high-frequency trading assets to capture rapid changes, while longer segments are used for low-frequency holding assets to avoid noise interference. Optionally, the value indicator can be standardized in terms of dimensions and fixed offsets caused by known non-market factors can be eliminated during the extraction stage to ensure the consistency of the difference calculation. Optionally, a sliding confidence interval correction is introduced during the comparison process, allowing the normal market volatility range parameter to be adaptively fine-tuned according to the recent data distribution. It can be understood that the basic value trend line has already filtered out the impact of extreme events during the construction stage, so the difference calculation focuses more on potential abnormal signals rather than occasional disturbances.

[0071] See Figure 3In one embodiment of the present invention, for each identified anomalous asset, user interaction behavior characteristics surrounding the anomalous asset are extracted starting from the time of identification. These characteristics include access frequency, holder change chain, and intensity of associated social topics. Access frequency refers to the number of times the digital asset is viewed, clicked, or accessed within a unit of time. The holder change chain records the time sequence of asset ownership transfers and related user identifiers. The intensity of associated social topics is quantified by analyzing the sentiment and discussion intensity of texts co-occurring with keywords related to the asset in public chat channels and forum posts within the metaverse. In a specific implementation, the extracted data from these three dimensions—access frequency, holder change chain, and intensity of associated social topics—are normalized and combined to form a three-dimensional initial behavioral feature vector. This initial behavioral feature vector represents a snapshot of user behavior at the time of identification of the anomalous asset. In subsequent consecutive time windows, even if the value indicators of the abnormal assets return to within the normal market fluctuation range, the tracking program continues to extract behavioral features of the same dimension according to a fixed sampling period. That is, it continuously collects the latest data on access frequency, holder change chain, and the strength of related social topics. Each sampling generates a new three-dimensional behavioral feature vector. These vectors are arranged in chronological order to form a behavioral feature vector sequence. The behavioral feature vector sequence fully records the dynamic evolution of user behavior of the abnormal assets from being marked to the subsequent observation period.

[0072] In practice, for each feature dimension in the behavioral feature vector sequence, the trajectory of its value change over time is calculated. The behavioral feature vector sequence includes a subsequence of access frequency dimension, a subsequence of holder change chain dimension, and a subsequence of related social topic strength dimension. For each of these three subsequences, with time as the horizontal axis and the normalized value of the dimension as the vertical axis, the sampling points are connected by linear interpolation to generate three independent behavioral feature change trajectory lines. These trajectory lines reflect the continuous change trend of a single behavioral feature dimension during the observation period. Smoothing and intensity accumulation calculations are performed on all behavioral feature change trajectory lines. The smoothing process uses a moving average filter to suppress short-term noise. The intensity accumulation calculation involves weighted summation of the values ​​of the three smoothed behavioral feature change trajectory lines at the same moment. The weighting coefficients can be understood as being allocated according to the indicative strength of different behavioral features to the abnormal state of the asset. By integrating the weighted sum at each moment throughout the observation period, an asset attention curve that comprehensively represents the sustained popularity of the behavior is obtained. The peak and width of the asset attention curve reflect the concentration and duration of the abnormal behavior pattern. In some embodiments, when generating behavioral feature change trajectory lines, for the holder change chain dimension, its value can be quantified as the number of ownership transfers per unit time, and the strength of associated social topics can be quantified as the product of the frequency of positive sentiment keywords and the total amount of discussion. In some embodiments, the sliding window length used in the smoothing process can be consistent with the length of the data segment in the dynamic value deviation detection to ensure the consistency of temporal processing granularity. Optionally, the weighted sum in the intensity accumulation calculation can be expressed by the following formula:

[0073]

[0074] in: Indicates at time The intensity of asset attention, Dimension index representing behavioral characteristics Indicates the first The pre-defined weight coefficients for each behavioral feature dimension Indicates the first The trajectory of behavioral feature changes for each behavioral feature dimension after smoothing at time [time] The numerical value. Optionally, before the initial behavioral feature vector is formed, the original data for each dimension can be normalized to eliminate the influence of dimensions and facilitate subsequent calculations. It can be understood that continuously tracking assets even when their value indicators have returned to normal helps to capture potentially lagging or hidden market manipulation and speculative behavior patterns. The asset attention curve integrates multi-dimensional behavioral characteristics, providing a more comprehensive portrayal of users' abnormal attention to assets compared to a single indicator.

[0075] In one embodiment of the present invention, the synchronicity coefficient of the changing trends of any two different behavioral feature dimensions at different points in time is analyzed from the behavioral feature vector sequence. The behavioral feature vector sequence consists of time-series data of three dimensions: access frequency, holder change chain, and strength of associated social topics. The synchronicity coefficient is obtained by calculating the consistency of the direction of change and the degree of correlation of the magnitude of the changes of two behavioral feature dimensions in the same time series. It can be expressed as the Pearson correlation coefficient or an equivalent measure to reflect whether the two dimensions show the same or opposite linkage within a specific time period. In a specific implementation, the synchronicity coefficients between all pairs of behavioral feature dimensions are arranged in the order of the feature dimensions to construct an initial behavioral synchronization matrix. The rows and columns of the initial behavioral synchronization matrix correspond to the three behavioral feature dimensions. The element in the i-th row and j-th column of the matrix is ​​the synchronicity coefficient value of the i-th dimension and the j-th dimension in the corresponding time period. The diagonal elements are set to fixed values ​​to indicate that the same dimension is completely synchronized. In practical implementation, the initial behavior synchronization matrix is ​​treated as a discrete observation result on the time axis. This is further used to generate an asset attention curve and a user behavior collaboration pattern matrix that evolve over time. To achieve this, the initial behavior synchronization matrix undergoes a time-dimensional moving average, calculating the mean of the synchronization coefficient within each moving window to form a dynamic behavior synchronization matrix sequence. In the actual implementation, the corresponding elements of all matrices in the dynamic behavior collaboration matrix sequence are fused to ultimately generate a user behavior collaboration pattern matrix reflecting complex user behavior association patterns. The fusion method involves accumulating and averaging all the means of each matrix position on the time axis, ensuring that the matrix elements reflect the strength of long-term stable collaboration relationships.

[0076] In practice, the initial behavioral synchronization matrix is ​​subjected to a time-dimensional moving average, and the mean of the synchronization coefficients within each moving window is calculated to form a dynamic behavioral synchronization matrix sequence. A fixed-length time moving window is set, and the initial behavioral synchronization matrix is ​​divided into multiple time segments according to timestamp order. Each time segment covers an equal-length period and contains several consecutive initial behavioral synchronization matrices. All initial behavioral synchronization matrices within each time segment are aligned and superimposed according to their element positions, and the arithmetic mean of all synchronization coefficients at each matrix position is calculated. This average represents the average degree of coordination at that position within the corresponding time window. The calculated arithmetic mean is then subjected to matrix reconstruction to generate the average behavioral synchronization matrix corresponding to the current time moving window. The average behavioral synchronization matrix retains the same row and column structure as the initial behavioral synchronization matrix, but each element is the mean within the time window. The average behavioral synchronization matrices corresponding to each time moving window are arranged and combined in chronological order to form a time-evolving dynamic behavioral synchronization matrix sequence. This dynamic behavioral synchronization matrix sequence can present the temporal changes in the coordination relationships between behavioral feature dimensions. In some embodiments, the length of the time sliding window can be selected based on the sampling period of the asset attention curve, so that the window division matches the temporal resolution of the behavioral feature vector sequence. In some embodiments, the initial behavioral synchronization matrix can limit the time lag range when calculating the synchronization coefficient, analyzing only the correlation at moments of the same direction or allowing a certain time shift, in order to capture lead-lag type collaborative patterns. Optionally, the arithmetic mean of all synchronization coefficients at each matrix position can be expressed by the following formula:

[0077]

[0078] in: Indicates the first Within the first time sliding window Line 1 The average synchronicity coefficient of the column This indicates the number of initial behavior synchronization matrices included within the time sliding window. Represents timestamp The corresponding initial behavior synchronization matrix in the first Line 1 The synchronization coefficient values ​​of the columns. Optionally, the arithmetic mean can be clipped before matrix reconstruction to limit all element values ​​to the valid range of the synchronization coefficient, avoiding abnormal mean values ​​from affecting subsequent fusion results. It can be understood that aligning and stacking the matrix according to its position and calculating the mean can eliminate the interference of single-point noise on the judgment of collaborative relationships, highlighting the overall trend. It can also be understood that the temporal arrangement of the dynamic behavior synchronization matrix sequence provides a continuous and stable input basis for the subsequent generation of the user behavior collaborative pattern matrix, enabling the fusion process to integrate collaborative strength information from different time periods to form a global relational view.

[0079] In one embodiment of the present invention, a model input interface is constructed to receive an asset attention curve and a user behavior collaboration pattern matrix, and the user behavior collaboration pattern matrix is ​​used as the initial weights of the edges between nodes in the graph. The model input interface includes a data adaptation and standardization module. The asset attention curve, as a time series, is converted into a fixed-dimensional feature vector. The element values ​​of the user behavior collaboration pattern matrix are scaled and directly mapped to the initial weight values ​​of the corresponding edge type in the heterogeneous graph. In a specific implementation, an asset-behavior heterogeneous graph is constructed, with digital assets and user behavior features as two types of nodes. Digital asset nodes represent monitored individuals with abnormal value, while user behavior feature nodes represent dimensions such as access frequency, holder change chain, and strength of related social topics. The association between assets and behaviors, as well as the collaboration relationship between behaviors, are used as edges. The edge between an asset node and a behavior feature node indicates that the asset has an observation value in that behavior dimension, and the edge weight can be initialized by the intensity component of the corresponding dimension in the asset attention curve. The edge between behavior feature nodes represents the collaboration relationship between different behavior dimensions, and the edge weight is directly assigned by the collaboration coefficient at the corresponding position in the user behavior collaboration pattern matrix. For the node and edge types of the asset-behavior heterogeneous graph, please refer to Table 1:

[0080] Table 1: Node and Edge Types of Asset-Behavior Heterogeneous Graph Asset Nodes Asset_i Temporal aggregation features extracted from asset attention curves Behavior Nodes Behavior_k Identifiers corresponding to specific behavioral dimensions (such as access frequency). edge type Connection Source of weight Asset-Behavior Correlation between assets and behavior The intensity of the dimension corresponding to the asset attention curve Behavior-Behavior Coordination between behavioral dimensions The corresponding position of the user behavior collaboration model matrix is ​​the collaboration coefficient.

[0081] In practical implementation, a graph attention network layer is constructed to perform multiple rounds of information aggregation and updating on nodes and edges in the asset-behavior heterogeneous graph, learning high-order embedding representations of nodes. The workflow of the graph attention network layer includes: for each target node in the asset-behavior heterogeneous graph, calculating the vector representation of all its neighboring node features after linear transformation, implemented using a trainable weight matrix; calculating the attention coefficients between the target node and the feature vectors of each neighboring node, where the attention coefficients are determined by the node features and the edge weights, where node features refer to the feature vector of the current node after linear transformation, and edge weights refer to the initial weights obtained from the heterogeneous graph structure or the weights updated in the previous iteration; processing the calculated attention coefficients through a normalization function to obtain standardized attention weights, typically using a softmax function applied to the set of attention coefficients of all neighbors of the target node; and finally, using the standardized attention weights to perform a weighted summation of the feature vectors of neighboring nodes, combined with the target node's own features, and passing it through a non-linear activation function to generate a new feature representation of the target node, completing one round of information aggregation. In practice, the graph attention network layer incorporates a multi-head attention mechanism, which executes the above process in parallel multiple times and concatenates or averages the outputs of different heads to stabilize the learning process and capture richer neighborhood information. Through multiple iterations of the graph attention network layer, asset nodes and behavior nodes can fuse information from multi-hop neighbors, ultimately obtaining a high-order embedding representation that includes high-order structural semantics and relationships.

[0082] In specific implementation, a graph structure readout and decision layer is constructed. The updated asset node embedding representations are pooled and mapped to the state anomaly probability space through a multilayer perceptron, outputting a state anomaly probability distribution map. The graph structure readout operation generates a global representation vector for the final set of embedding representations of all asset nodes using global average pooling or summation pooling. Optionally, the asset node embedding representations can be operated on directly without global pooling, preserving the independent representation of each asset. In specific implementation, the pooled vector or the independent embedding representation of each asset is input into a multilayer perceptron. The multilayer perceptron consists of alternating fully connected layers and activation functions, with the output dimension of its last layer matching the preset number of state evaluation dimensions. After the output of the multilayer perceptron is processed by the softmax function, a state anomaly probability distribution map of the target digital asset across multiple evaluation dimensions is obtained. The state anomaly probability distribution map is presented in the form of a matrix or vector set, where each element represents the probability that the corresponding asset is in an anomalous state across a specific evaluation dimension. In some embodiments, the attention coefficient can be calculated using the following formula:

[0083]

[0084] in: Represents the target node Its neighboring nodes The original attention coefficients between them This represents a trainable parameter vector. This represents the shared linear transformation weight matrix applied to node features. and Representing the target node respectively and neighboring nodes The input feature vector, This represents a vector concatenation operation. Representing nodes in an asset-behavior heterogeneous graph and The weight of the edges between them. This represents a non-linear activation function. In some embodiments, the number of layers in the graph attention network can be set to 2 to 3 to ensure that the model can capture sufficient neighborhood-wide information without becoming overly smooth. Optionally, during information aggregation, the target node's own features can be transformed by an independent, trainable self-connected weight matrix before being added to the weighted features of its neighbors. Optionally, the edge weights... In the iterative process of the graph attention network layer, it can participate in attention calculation as a learnable scaling factor, rather than remaining fixed. It can be understood that by introducing edge weights to calculate the attention coefficients, the model can distinguish the differences in the impact of different types and strengths of associations on information propagation. It can also be understood that the graph structure readout and decision layer transform the complex learning results of the graph structure into an intuitive probability distribution output, providing direct quantitative evidence for the final management strategy determination.

[0085] See Figure 4 This is a heatmap of user behavior collaboration patterns, used to demonstrate the strength of collaborative relationships among four core user behavior dimensions in the Metaverse digital asset management scenario. This map helps Metaverse digital asset managers identify patterns in user behavior linkages. For example, when abnormal changes in the holder change chain are detected, the focus can be on the synchronized changes in social topic intensity and trading activity, providing early warnings of abnormal asset fluctuations. For behaviors with low collaboration, operational strategies can be designed to strengthen their correlation and improve asset liquidity. The heatmap intuitively quantifies the strength of relationships between different user behaviors. For instance, the high collaboration between "holder change chain," "social topic intensity," and "trading activity" indicates that changes in asset ownership are the core behavior driving social discussion and trading, providing a clear quantitative basis for understanding user behavior logic.

[0086] In one embodiment of the present invention, an abnormal state probability distribution map is analyzed. This map is output by an asset state analysis model based on a graph neural network. It can be a two-dimensional matrix, where rows represent different digital assets, columns represent different evaluation dimensions, and matrix element values ​​represent the probability of an abnormal state for the corresponding asset in the corresponding evaluation dimension. Continuous regions in the abnormal state probability distribution map whose probability values ​​exceed a set threshold are identified. This threshold is predefined by the system administrator based on risk tolerance. A continuous region refers to a connected area on the evaluation dimension-asset plane composed of adjacent cells whose probability values ​​all exceed the threshold. In a specific implementation, continuous regions are defined as high-risk clusters. High-risk clusters are aggregates of abnormal signals in the abnormal state probability distribution map. Each high-risk cluster may cover one or more digital assets with high-probability abnormal states in one or more evaluation dimensions. Core indicators for each high-risk cluster are calculated. These core indicators include the duration of the abnormal state, the number of associated user behavior feature dimensions, and the average abnormal probability intensity within the cluster. The duration of the abnormal state is calculated by tracing the earliest and latest occurrence times of the cells constituting the high-risk cluster in the time series. The statistical analysis of the number of associated user behavior feature dimensions covers the types of evaluation dimensions covered by the high-anomaly-risk cluster. These evaluation dimensions are directly related to user behavior features. The average anomaly probability intensity within the cluster is calculated by summing the probability values ​​of all cells within the high-anomaly-risk cluster and dividing by the total number of cells.

[0087] In implementation, a management strategy level mapping rule base is established. This rule base is a predefined lookup table or rule set, defining management strategy levels corresponding to different combinations of core indicators. A core indicator combination is an ordered triple consisting of three indicators: the duration of the abnormal state, the number of associated user behavior feature dimensions, and the average anomaly probability intensity within the cluster. Each triple maps to a discrete management strategy level, which can include intervention levels ranging from low to high, such as observation, alerts, transaction restrictions, and freezes. In practice, the core indicators of each high-risk cluster are matched against the management strategy level mapping rule base, assigning a candidate management strategy level to each cluster. The matching process involves precisely or range-wise matching the core indicator triples of high-risk clusters with entries in the rule base to find the corresponding management strategy level. Among all the candidate management strategy levels assigned to high-risk clusters, the final management strategy level applied to the digital asset is determined according to predefined conflict resolution rules. These conflict resolution rules handle situations where the same digital asset is assigned multiple candidate management strategy levels because it belongs to different high-risk clusters.

[0088] In some embodiments, the average anomaly probability intensity within a cluster can be expressed by the following formula:

[0089]

[0090] in: Indicates a cluster with high anomaly risk The average anomaly probability intensity within the cluster, Indicates a cluster with high anomaly risk The total number of cells contained. The row index in the probability distribution diagram of abnormal states is... Column index is cell coordinates, This represents the probability value of the abnormal state in the cell. In some embodiments, the preset conflict resolution rule can adopt the highest-level priority principle, that is, when the same digital asset appears in multiple high-risk clusters, the final management strategy level is the highest among all associated candidate management strategy levels. Optionally, the definition of the management strategy level mapping rule base can be based on the analysis of historical handling cases, associating the core indicator combination before the incident with the management strategy level that was verified to be effective after the incident with asset cases that have experienced risk events in the past and have taken specific intervention measures, thereby forming mapping entries. Optionally, the threshold can be set to a dynamically adjustable value, such as taking a certain high percentile of all probability values ​​in the abnormal state probability distribution map, so that the identification of high-risk clusters can adapt to the overall probability distribution. It can be understood that identifying risk clusters through the clustering pattern of probability values ​​rather than isolated high points can reduce the risk of misjudging occasional noise signals as abnormal.

[0091] See Figure 5 This is a trend curve chart showing the change in attention to digital assets in the metaverse over time, possessing multi-dimensional practical value in digital asset management scenarios. After the peak of attention, the curve experiences a sharp drop. This "boom and bust" pattern is often highly correlated with asset speculation and short-term speculative behavior, and can be considered a typical characteristic of high-risk assets. When attention falls below the normal baseline or exceeds the historical peak, a risk warning can be triggered, and the cause of the anomaly can be further verified by combining a user behavior coordination matrix. Successful experiences from the peak of attention can be reviewed and replicated in the operation of other assets; for the subsequent decline in attention, retention strategies can be designed to extend the asset's popularity. By comparing the attention curve shapes of different assets, "healthy volatility" and "speculative volatility" assets can be distinguished, providing a basis for asset classification management.

[0092] 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 management method based on the metaverse, characterized in that, The method includes: Historical digital asset interaction records are extracted from multi-source metaverse space data, and asset type classification and basic value modeling are performed on them to generate basic value curves and normal market fluctuation range parameter sets corresponding to different asset types. Receive real-time data streams of target digital asset transactions and usage within the metaverse, and perform dynamic value deviation detection on the target digital asset transaction and usage data streams based on the basic value curve and the normal market fluctuation range parameter set. Mark digital assets whose value fluctuations exceed the corresponding normal market fluctuation range parameters to obtain assets with abnormal value. The user behavior trajectory of the aforementioned abnormal value assets is tracked and aggregated within a continuous time window to generate an asset attention curve and a user behavior collaborative pattern matrix that evolves over time. The asset attention curve and user behavior collaborative pattern matrix are input into the asset status analysis model based on graph neural network. The asset status analysis model outputs the probability distribution map of the target digital asset's status anomaly across multiple evaluation dimensions. Based on the clustering pattern and intensity of probability values ​​in the aforementioned abnormal state probability distribution diagram, the final management strategy level of the target digital asset is determined.

2. The digital asset management method based on the metaverse as described in claim 1, characterized in that, The asset type classification and basic value modeling for it include: Digital assets in the metaverse are divided into multiple asset classes with different economic attributes using predefined metadata standards; For each asset class, transaction and usage data of the corresponding asset class under normal market conditions are extracted from the historical digital asset interaction records; Value characteristic analysis and volatility statistics are performed on the transaction and usage data of the corresponding asset classes under normal market conditions to establish a basic value trend line and market volatility statistical distribution chart for each asset class; Based on the aforementioned fundamental value trend line and market volatility statistical distribution chart, the normal market volatility range parameter for each asset class is defined.

3. The digital asset management method based on the metaverse as described in claim 2, characterized in that, The dynamic value deviation detection of the target digital asset transaction and usage data stream based on the basic value curve and the normal market fluctuation range parameter set includes: The real-time collected data stream of target digital asset transactions and usage within the metaverse is divided into continuous data segments in chronological order. For each data segment, the difference between its value index and the basic value trend line is calculated under the corresponding asset class to obtain the absolute value deviation. Simultaneously, the rate of change of value of the current data segment compared to the previous data segment within the corresponding asset class is calculated to obtain the value volatility. The absolute deviation of value and the volatility of value are compared with the normal market volatility range parameters defined in the market volatility statistical distribution chart, respectively. If either the absolute deviation of value or the volatility of value continuously exceeds the corresponding normal market fluctuation range parameter for a preset duration, the current asset will be identified as an asset with abnormal value.

4. The digital asset management method based on the metaverse as described in claim 3, characterized in that, The step of tracking and aggregating user behavior trajectories within a continuous time window for the assets with abnormal value includes: For each identified anomalous asset, starting from the time of its identification, user interaction behavior features surrounding the anomalous asset are extracted, including access frequency, holder change chain, and intensity of associated social topics, to form an initial behavioral feature vector. In subsequent consecutive time windows, even if the value indicators of the abnormal assets return to within the normal market fluctuation range parameters, the same behavioral characteristics will continue to be tracked and extracted to form a sequence of behavioral characteristic vectors. For each feature dimension in the behavioral feature vector sequence, calculate the trajectory of its value change over time to generate multiple behavioral feature change trajectory lines; By smoothing and intensity accumulation calculations on all behavioral characteristic change trajectories, the asset attention curve representing the sustained popularity of the behavior is obtained.

5. The digital asset management method based on the metaverse as described in claim 4, characterized in that, The generated asset attention curve and user behavior collaborative pattern matrix that evolve over time include: From the sequence of behavioral feature vectors, analyze the synchronicity coefficient of the changing trends of any two different behavioral feature dimensions at different points in time; Arrange the synchronization coefficients between all pairs of behavioral feature dimensions in the order of the feature dimensions to construct the initial behavioral synchronization matrix; The initial behavior synchronization matrix is ​​subjected to a time-dimensional moving average, and the mean of the synchronization coefficient within each moving window is calculated to form a dynamic behavior synchronization matrix sequence. The corresponding elements of all matrices in the dynamic behavior coordination matrix sequence are merged to finally generate the user behavior coordination pattern matrix that reflects the complex correlation pattern of user behavior.

6. The digital asset management method based on the metaverse as described in claim 1, characterized in that, The construction and operation of the asset status analysis model based on graph neural networks include: A model input interface is constructed to receive the asset attention curve and the user behavior collaboration pattern matrix, and the user behavior collaboration pattern matrix is ​​used as the initial weight of the edges between nodes in the graph. Construct an asset-behavior heterogeneous graph, with digital assets and user behavior features as two types of nodes, and the association between assets and behaviors and the synergistic relationship between behaviors as edges; A graph attention network layer is constructed to perform multi-round information aggregation and updating on nodes and edges in the asset-behavior heterogeneous graph, and to learn the high-order embedding representation of nodes. A graph structure readout and decision layer is constructed, the updated asset node embedded representation is pooled, and mapped to the state anomaly probability space through a multilayer perceptron, outputting the state anomaly probability distribution map.

7. A digital asset management method based on a metaverse as described in claim 6, characterized in that, The workflow of the graph attention network layer includes: For each target node in the asset-behavior heterogeneous graph, calculate the vector representation of the features of all its neighboring nodes after linear transformation; Calculate the attention coefficient between the target node and the feature vector of each neighbor node. The attention coefficient is determined by the node features and the weights of the connecting edges. The calculated attention coefficients are processed by a normalization function to obtain standardized attention weights; The standardized attention weights are used to perform a weighted summation of the feature vectors of neighboring nodes, combined with the target node's own features, and then processed through a nonlinear activation function to generate a new feature representation of the target node.

8. A digital asset management method based on a metaverse as described in claim 1, characterized in that, The final management strategy level for determining the target digital asset includes: Analyze the state anomaly probability distribution map, identify continuous regions in the map where the probability value exceeds a set threshold, and define the continuous regions as high anomaly risk clusters; Calculate the core metrics for each high-risk cluster, including the duration of the abnormal state, the number of associated user behavior feature dimensions, and the average anomaly probability intensity within the cluster. A management strategy level mapping rule base is established, in which management strategy levels are defined for different combinations of core indicators; The core indicators of each high-risk cluster are matched with the management strategy level mapping rule base, and a candidate management strategy level is assigned to each cluster; Among all the candidate management strategy levels assigned to high-risk clusters, the final management strategy level applied to the digital asset is determined according to the preset conflict resolution rules.

9. A digital asset management method based on a metaverse as described in claim 5, characterized in that, The step of performing a time-dimensional moving average on the initial behavior synchronization matrix to calculate the mean of the synchronization coefficients within each moving window, forming a dynamic behavior synchronization matrix sequence, includes: A fixed-length time sliding window is set, and the initial behavior synchronization matrix is ​​divided into multiple time segments according to the timestamp order; All initial behavior synchronization matrices within each time segment are aligned and superimposed according to the matrix element positions, and the arithmetic mean of all synchronization coefficients at each matrix position is calculated. The calculated arithmetic mean is subjected to matrix reconstruction to generate the average behavior synchronization matrix corresponding to the current time sliding window. The average behavior synchronization matrices corresponding to each time sliding window are arranged and combined in chronological order to form a dynamic behavior synchronization matrix sequence that evolves over time.

10. A digital asset management system based on a metaverse, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the digital asset management method based on the metaverse as described in any one of claims 1 to 9.

Images