Distributed multi-source digital content tamper-proof hash evidence storage method based on block chain

Abstract

The invention provides a distributed multi-source digital content tamper-proof Hash evidence storage method based on a block chain, and relates to the technical field of content tamper-proof, and the method comprises the steps: standardizing multi-source digital content, constructing an evolution pedigree tree, implanting a detector to capture change features, and tracking a propagation path; constructing a dynamic perceptual hash network to generate a dynamic hash value containing associated information; and verifying and generating an evidence storage record through a progressive consensus mechanism. According to the invention, efficient tamper-proof evidence storage of the multi-source digital content is realized, and the data traceability and verification efficiency are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of content tamper-proofing, and in particular to a distributed multi-source digital content tamper-proofing hash storage method based on a blockchain. BACKGROUND

[0002] With the rapid development of digital content, various information sources and transmission channels are increasingly rich, and the authenticity and integrity protection of multi-source digital content has become a problem to be solved. The prior art stores and protects the digital content by a blockchain storage method, generates a content characteristic value by using a hash algorithm, and records it on the blockchain to realize content verification and tracing.

[0003] However, the prior art still has deficiencies, the traditional hash algorithm can only reflect the static characteristics of the content, and cannot effectively describe the correlation between the multi-source digital contents and the evolution characteristics; the tracking mechanism for the content transmission path and the change behavior is lacking in the blockchain storage process, and it is difficult to accurately identify content tampering and traceability; the verification nodes use a unified consensus mechanism for hash value verification, without considering the hierarchical relationship between the nodes, which easily causes low verification efficiency and consensus delay.

[0004] In summary, the present application aims to solve the above technical problems, and proposes a distributed multi-source digital content tamper-proofing hash storage method based on a blockchain, which traces the content change characteristics and transmission path by constructing an evolution pedigree tree, perceives the state of the associated content by using a graph neural network, and generates a dynamic hash value, uses a hierarchical verification mechanism based on node coverage to improve the consensus efficiency, and realizes the reliable storage and effective tamper-proofing of the multi-source digital content. SUMMARY

[0005] The present application provides a distributed multi-source digital content tamper-proofing hash storage method based on a blockchain, which can solve the problems in the prior art.

[0006] The first aspect of the embodiment of the present application is,

[0007] A distributed multi-source digital content tamper-proofing hash storage method based on a blockchain is provided, comprising:

[0008] Receiving multi-source digital content, converting the multi-source digital content into standardized data;

[0009] Constructing an evolution pedigree tree of the standardized data, implanting a detector at a branch node, capturing change characteristics and tracing transmission paths, calculating the correlation strength value between the data, and constructing a content correlation network;

[0010] Based on the content correlation network, a dynamic perception hash network is constructed, the state change of the associated content is perceived by using a graph neural network, and a dynamic hash value containing associated information is generated for each content node.

[0011] The node coverage degree is calculated based on the evolutionary phylogenetic tree in the blockchain verification pool to form a hierarchical verification group, and a dynamic hash value is verified through a progressive consensus mechanism, and the verification result is transmitted layer by layer to generate a storage record containing the dynamic hash value, the evolutionary phylogenetic tree information and the timestamp;

[0012] When receiving a query request, the storage record is obtained and the evolutionary phylogenetic tree is reconstructed, the hash value is calculated and verified, and the verification result is returned.

[0013] In an optional embodiment,

[0014] An evolutionary phylogenetic tree of the standardized data is constructed, a probe is implanted at a branch node to capture a change feature and track a propagation path, a correlation strength value between data is calculated, and a content correlation network is constructed, including:

[0015] The standardized data of the multi-source digital content is taken as a root node, an evolutionary phylogenetic tree is constructed based on iterative versions of the standardized data, and a branch node is constructed by comparing feature differences and time intervals of the standardized data in different iterative versions;

[0016] A probe is implanted at the branch node, the probe monitors and records a feature change amount and a propagation time interval of the standardized data passing through the corresponding branch node in real time, and a change feature of the branch node is generated;

[0017] A feature diffusion strength function is constructed based on the change feature, a feature diffusion range of each branch node is calculated, a propagation path is determined according to an overlap degree of the feature diffusion range, and a feature diffusion cumulative distance between adjacent nodes on the propagation path is calculated;

[0018] The feature diffusion cumulative distance is taken as a correlation strength value between branch nodes, and the correlation strength value exponentially decays with an increase of the feature diffusion cumulative distance;

[0019] The branch node is taken as a network node, and a correlation strength value higher than a preset strength threshold is taken as a weight of a connection edge, and a content correlation network of the standardized data is constructed.

[0020] In an optional embodiment,

[0021] A feature diffusion strength function is constructed based on the change feature, a feature diffusion range of each branch node is calculated, a propagation path is determined according to an overlap degree of the feature diffusion range, and a feature diffusion cumulative distance between adjacent nodes on the propagation path is calculated, including:

[0022] A feature diffusion strength function is established based on a timestamp and a node spatial position of the change feature, the feature diffusion strength function contains a time decay term and a spatial decay term, and a feature diffusion range of each branch node is generated;

[0023] According to the overlapping area of the feature diffusion range between branch nodes, a propagation matrix is established, and when the overlapping degree of the feature diffusion ranges of two branch nodes exceeds a preset diffusion threshold, a propagation link between the nodes is recorded at the corresponding position of the propagation matrix;

[0024] Based on the propagation link, the branch nodes are concatenated into a propagation sequence, and the time sequence continuity of the nodes in the propagation sequence is detected;

[0025] The propagation sequence that meets the time sequence constraint and has no loop is determined as a propagation path, and the cumulative distance of feature diffusion between adjacent nodes on the propagation path is calculated.

[0026] In an optional embodiment,

[0027] Based on the content association network, a dynamic perception hash network is constructed, and a graph neural network is used to perceive the state change of the associated content, and a dynamic hash value containing associated information is generated for each content node.

[0028] Based on the content features of the nodes in the content association network, an initial state vector of the node is constructed, and the association strength features of the nodes are calculated according to the connection edge weights between adjacent nodes; the content features and the association strength features are combined to construct a node state vector; the node state vectors of adjacent nodes and the connection edge weights are combined and encoded to obtain association edge features, which reflect the change of the association strength between node pairs;

[0029] The shortest path distance between nodes is calculated using a spatial propagation function and is divided into multiple levels, the node state vectors in each level are weighted and fused based on the exponential decay coefficient of the level number and the normalized weight of the association edge feature strength, and the multi-scale neighborhood representation of the node is generated by combining scale adaptation and residual mapping; the multi-scale neighborhood representation and the historical state information of the node are fused by using a recursive state transition unit to generate the spatio-temporal fusion features of the node;

[0030] The hash features are obtained by nonlinear transformation of the spatio-temporal fusion features, and the initial hash values are obtained by binary quantization of the hash features; the propagation influence range of the state change of the node is calculated based on the node association topology structure, and the initial hash values are dynamically adjusted according to the state change of the associated nodes in the propagation influence range to obtain dynamic hash values that can adaptively reflect the state change of the node.

[0031] In an optional embodiment,

[0032] The shortest path distance between nodes is calculated by using a spatial propagation function, and multi-level division is performed. The state vectors of the nodes in each level are weighted and fused based on the exponential decay coefficient of the level sequence number and the normalized weight of the correlation edge feature strength. A multi-scale neighborhood representation of the node is generated by combining scale adaptation and residual mapping, including:

[0033] The shortest path distance between nodes is calculated by using a spatial propagation function, and multi-level division is performed. The state vectors of the nodes in each level are weighted and fused based on the exponential decay coefficient of the level sequence number and the normalized weight of the correlation edge feature strength. A multi-scale neighborhood representation of the node is generated by combining scale adaptation and residual mapping, including:

[0034] The spatial propagation function calculates the distance decay coefficient corresponding to each level according to the level sequence number, and makes the distance decay coefficient decrease with the increase of the level through an exponential decay function. For each target node in each level, the correlation strength weight is calculated based on the correlation edge feature strength between the target node and the center node, and the correlation strength weight is obtained by normalizing the correlation edge feature strength in the level.

[0035] The spatial propagation function multiplies the state vector of each node in each level by the distance decay coefficient corresponding to the level, and multiplies the corresponding correlation strength weight of the node and sums it in the level to generate the node feature of the level. The node features of all levels are concatenated in level order, and an initial multi-scale neighborhood representation is generated through nonlinear transformation.

[0036] The scale adaptation weight is calculated for the node features of each level, and the scale adaptation weight is used to weight and fuse the node features of each level. The residual mapping is performed on the initial multi-scale neighborhood representation, and the mapping result is added to the weighted and fused features to generate an optimized multi-scale neighborhood representation.

[0037] In an alternative embodiment,

[0038] In a blockchain verification pool, the node coverage is calculated based on the evolutionary pedigree tree to form a hierarchical verification group, and the dynamic hash value is verified by a progressive consensus mechanism. The verification result is transmitted layer by layer to generate a storage record containing the dynamic hash value, the evolutionary pedigree tree information and the timestamp, including:

[0039] The distance value between nodes is calculated based on the evolutionary pedigree tree in the blockchain verification pool, and the distance value between nodes is dynamically adjusted and weighted calculated based on the node reputation and the adaptive decay factor to obtain the local coverage of the node. The global coverage of the node is calculated based on the ratio of the local coverage to the sum of the local coverages of all nodes in the verification pool.

[0040] The reference threshold value and the exponential operation result of the hierarchical attenuation rate are taken as a hierarchical threshold value, the nodes are layered according to the hierarchical threshold value, and the nodes with a global coverage greater than the corresponding hierarchical threshold value are divided into a verification group of a corresponding level;

[0041] The nodes in each verification group verify the dynamic hash value and generate a verification signature, a ratio of the number of verification signatures in the verification group to the total number of nodes in the verification group is taken as a consensus degree of the verification group, and when the consensus degree of the verification group is greater than a preset consensus threshold value, a verification result is transmitted to an upper-level verification group;

[0042] The dynamic hash value, the evolutionary pedigree tree information, the timestamp and the verification signature of each verification group are integrated to generate a block chain storage record.

[0043] In an optional embodiment,

[0044] The distance value between nodes is calculated based on the evolutionary pedigree tree in the block chain verification pool, the distance value between nodes is dynamically adjusted and weighted calculated based on the node reputation and the adaptive attenuation factor, and the local coverage of nodes is obtained, including:

[0045] The distance value between nodes is calculated based on the evolutionary pedigree tree in the block chain verification pool, the distance value between nodes is dynamically adjusted and weighted calculated based on the node reputation and the adaptive attenuation factor, and the local coverage of nodes is obtained, including:

[0046] The adaptive attenuation factor is constructed, the adaptive attenuation factor is composed of a preset basic attenuation rate and network dynamic parameters, the network dynamic parameters include the current verification load of the block chain verification pool, the network congestion degree of the block chain verification pool and the node activity degree of the block chain verification pool, and the distance value between nodes is dynamically adjusted according to the adaptive attenuation factor.

[0047] The influence degree of the neighborhood nodes of the nodes in the block chain verification pool is set as a node weight coefficient based on the node reputation, the node weight coefficient and the dynamically adjusted distance value between nodes are weighted calculated, the evolutionary correlation strength between nodes is obtained, and the local coverage of nodes is obtained by accumulating the evolutionary correlation strength between nodes.

[0048] In an embodiment of the present invention, by converting multi-source digital content into standardized data and constructing an evolutionary pedigree tree, accurate tracking and tracing of digital content changes are achieved, the association relationship and propagation path between content can be effectively identified, the accuracy and reliability of digital content management are improved, and the problem that traditional methods are difficult to handle multi-source heterogeneous data is solved; a dynamic perception hash network constructed based on a content association network uses graph neural network technology to perceive the state changes of associated content, generate dynamic hash values ​​containing associated information, enhance anti-tampering capabilities, enable the system to intelligently identify content changes and retain change history, and improve the credibility and integrity of evidence; the use of a progressive consensus mechanism and a hierarchical verification group structure significantly improves verification efficiency and system scalability, reduces computing resource consumption, and at the same time ensures the immutability and traceability of evidence records through blockchain technology, providing a reliable legal guarantee and rights protection mechanism for digital content. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 Schematic diagram of a process of an embodiment of the present invention;

[0050] Figure 2 This is a comprehensive comparative analysis table of the multi-dimensional performance of the feature diffusion model;

[0051] Figure 3 This is a comparative analysis diagram of feature distribution before and after optimization of multi-scale neighborhood representation;

[0052] Figure 4 A radar chart showing the relationship between node reputation and verification accuracy;

[0053] Figure 5 This is a data table showing the relationship between node reputation and verification accuracy. DETAILED DESCRIPTION

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0055] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0056] Figure 1 This is a flow chart of a distributed multi-source digital content tamper-proof hash evidence storage method based on blockchain according to an embodiment of the present invention. Figure 1As shown, the method includes:

[0057] receiving multi-source digital content, and converting the multi-source digital content into standardized data;

[0058] Constructing an evolutionary pedigree tree of the standardized data, implanting detectors at branch nodes, capturing change features and tracing propagation paths, calculating correlation strength values ​​between data, and constructing a content association network;

[0059] Building a dynamic perception hash network based on the content association network, using a graph neural network to perceive the state changes of associated content, and generating a dynamic hash value containing associated information for each content node;

[0060] In the blockchain verification pool, a hierarchical verification group is formed based on the node coverage calculated based on the evolutionary pedigree tree. The dynamic hash value is verified through a progressive consensus mechanism. The verification results are passed layer by layer and a certificate record containing the dynamic hash value, evolutionary pedigree tree information and timestamp is generated.

[0061] When receiving a query request, it obtains the evidence record and reconstructs the evolutionary lineage tree, calculates the hash value for verification, and returns the verification result.

[0062] In an optional embodiment, constructing an evolutionary pedigree tree of the standardized data, implanting detectors at branch nodes, capturing change features and tracing propagation paths, calculating correlation strength values ​​between data, and constructing a content association network includes:

[0063] Taking the standardized data of multi-source digital content as the root node, constructing an evolutionary pedigree tree based on iterative versions of the standardized data, and constructing branch nodes by comparing the feature differences and time intervals of the standardized data in different iterative versions;

[0064] A detector is implanted in the branch node, and the detector monitors and records in real time the characteristic change amount and propagation time interval of the standardized data when it passes through the corresponding branch node, and generates a change characteristic of the branch node;

[0065] Constructing a feature diffusion intensity function based on the change feature, calculating the feature diffusion range of each branch node, determining a propagation path according to the degree of overlap of the feature diffusion range, and calculating the feature diffusion cumulative distance between adjacent nodes on the propagation path;

[0066] The feature diffusion cumulative distance is used as the association strength value between branch nodes, and the association strength value decays exponentially with the increase of the feature diffusion cumulative distance;

[0067] The branch nodes are used as network nodes, and the association strength values ​​higher than a preset strength threshold are used as weights of connection edges to construct a content association network of the standardized data.

[0068] In one specific implementation, taking the product description data of an e-commerce platform as an example, the initial version of the product description data (such as "2023 high-performance smartphone, 8GB+256GB, 5000mAh battery") is taken as the root node. The system collects the iterative versions of the product description at different time points, such as the first iterative version "2023 new high-performance smartphone, 8GB+256GB large memory, 5000mAh super large battery", and the second iterative version "2023 new high-performance smartphone, 8GB+256GB large memory, 5000mAh super large battery, supports fast charging technology". By comparing the text feature differences (such as the addition of the words "new", "large memory", "super large", "supports fast charging technology") and the time intervals (such as the interval between version one and the root node is 7 days, and the interval between version two and version one is 15 days) between these versions, the system constructs branch nodes. Each branch node represents a significant change in data, such as the description language enhancement from the root node to the first branch node, and the functional feature supplement from the first branch node to the second branch node.

[0069] A probe is implanted in the branch node, which is a piece of embedded program code that monitors and records the feature change amount and propagation time interval of standardized data passing through the corresponding branch node in real time. Taking the above product description as an example, when the second iterative version of the description is quoted by other merchants or forwarded by users, the probe will record the frequency of the feature word "fast charging technology" being quoted, the quoting time, and the transformation after quoting (such as becoming "super fast charging technology" or "flash charging technology"). Through these records, the probe generates the change features of the branch node, such as {"feature word": "fast charging technology", "citation frequency": 156, "average citation delay": 2.3 days, "transformation rate": 0.42}.

[0070] Based on the change features, a feature diffusion intensity function is constructed, which describes the variation of the intensity of feature diffusion from one node to the outside with time and space. Taking the "fast charging technology" feature as an example, its initial diffusion intensity is 100 (indicating complete retention of the original feature), and as time passes and the propagation distance increases, the diffusion intensity gradually decreases. When the feature is propagated to a third-party platform, its diffusion intensity may decrease to 65, indicating that the original feature retention is 65%. The system determines the propagation path by calculating the feature diffusion range of each branch node (such as the diffusion range of "fast charging technology" covering the mobile phone, tablet, and notebook subcategories in the electronic product category), and according to the overlap degree of the diffusion ranges of different nodes (such as the overlap degree of "fast charging technology" and "large memory" is 0.78).

[0071] The cumulative distance of feature diffusion between adjacent nodes on the propagation path is calculated, and the cumulative distance considers factors such as feature deformation degree, propagation time delay and spatial span. The cumulative distance of feature diffusion from node A ("fast charging technology") to node B ("super fast charging") is 0.35, indicating that there is a small feature deformation between them; and the cumulative distance from node A to node C ("long endurance") is 0.82, indicating that the feature correlation between them is weak.

[0072] The cumulative distance of feature diffusion is taken as the correlation strength value between branch nodes, and the correlation strength value decreases exponentially with the increase of the cumulative distance of feature diffusion. The correlation strength value can be represented as a value between 0 and 1. When the cumulative distance is 0, the correlation strength is 1 (complete correlation); when the cumulative distance increases, the correlation strength decreases rapidly. For example, the cumulative distance of 0.35 corresponds to a correlation strength value of 0.7, and the cumulative distance of 0.82 corresponds to a correlation strength value of only 0.2.

[0073] The branch nodes are taken as network nodes, and the correlation strength value higher than a preset strength threshold (such as 0.5) is taken as the weight of the connection edge to construct the content correlation network of the standardized data. If the threshold is set to 0.5, a connection edge with a weight of 0.7 will be established between node A and node B, and no connection will be established between node A and node C because the correlation strength value 0.2 is lower than the threshold.

[0074] The content correlation network constructed by the above method is visualized as a complex network structure, in which the nodes represent different change states of the data, the connection edges represent the correlation between the changes, and the thickness of the edges represents the correlation strength. For example, in the product description network of an e-commerce platform, a sub-network centered on "fast charging technology" may be formed, connecting "flash charging", "super fast charging", "charging speed" and other related feature nodes, and these nodes are connected to other feature nodes, forming a complex correlation network structure.

[0075] In an optional implementation, a feature diffusion strength function is constructed based on the change feature, the feature diffusion range of each branch node is calculated, the propagation path is determined according to the overlap degree of the feature diffusion range, and the cumulative distance of feature diffusion between adjacent nodes on the propagation path comprises:

[0076] Based on the timestamp and the spatial position of the change feature, a feature diffusion strength function is established, which includes a time decay term and a spatial decay term, and the feature diffusion range of each branch node is generated;

[0077] According to the overlapping area of the feature diffusion range between branch nodes, a propagation matrix is established. When the overlapping degree of the feature diffusion range of two branch nodes exceeds a preset diffusion threshold, the propagation link between the nodes is recorded at the corresponding position of the propagation matrix;

[0078] Connecting branch nodes in series into a propagation sequence based on the propagation link, and detecting the temporal continuity of nodes in the propagation sequence;

[0079] A propagation sequence that meets the timing constraint and has no loop is determined as a propagation path, and the characteristic diffusion cumulative distance between adjacent nodes on the propagation path is calculated.

[0080] In a specific embodiment, a feature diffusion intensity function is established based on the timestamp of the changed feature and the spatial position of the node. The function includes a time decay term and a spatial decay term, and is used to generate the feature diffusion range of each branch node.

[0081] For each branch node in the system, obtain its change feature information, including the timestamp T of the change and the node's spatial location coordinates P. The timestamp T records the time when the change occurred in seconds, for example, "2023-05-10 14:30:25" converted to a Unix timestamp; the spatial location coordinates P can be two-dimensional or three-dimensional coordinates, such as the location (10.5, 20.3) in the network topology.

[0082] The feature diffusion intensity function I(d, t) represents the diffusion intensity of a feature at a distance of d and a time difference of t. This function consists of a time decay term and a spatial decay term. The time decay term uses an exponential decay form, while the spatial decay term uses a Gaussian decay form. Specifically, for node i, its feature diffusion intensity can be expressed as a function of distance d and time difference t. As the distance or time difference increases, the diffusion intensity decreases.

[0083] The time decay coefficient α is set to 0.05, representing the decay rate per unit time; the spatial decay coefficient β is set to 0.1, representing the decay rate of spatial diffusion. When the feature diffusion intensity is greater than the preset threshold θ (for example, θ = 0.3), the point is considered to be within the feature diffusion range of the node.

[0084] For example, for node A (timestamp 1620642625, location coordinates (10.5, 20.3)) and node B (timestamp 1620642725, location coordinates (12.8, 19.6)), the characteristic diffusion intensity of node A is calculated 100 seconds later and at a distance of 3.1 units. Based on the time decay and spatial decay calculations, the characteristic diffusion intensity is approximately 0.42, which is greater than the threshold of 0.3. Therefore, this point is within the characteristic diffusion range of node A.

[0085] A propagation matrix is ​​established based on the overlapping area of ​​the feature diffusion range between branch nodes. When the overlap of the feature diffusion range of two branch nodes exceeds the preset diffusion threshold, the propagation link between the nodes is recorded at the corresponding position of the propagation matrix.

[0086] For each pair of nodes (i, j) in the system, calculate the overlap of their feature diffusion ranges. Sample multiple points on a spatial grid (e.g., a 10×10 grid) and count the number of points where the feature diffusion strength of both node i and node j is greater than a threshold θ. Divide this by the total number of sampled points to obtain the overlap R(i, j).

[0087] Set the overlap threshold γ (for example, γ = 0.25). When R(i, j)>γ, it is considered that there is a propagation link between node i and node j, and the corresponding position M[i, j] in the propagation matrix M is recorded as 1, otherwise it is recorded as 0.

[0088] For example, the calculated result of the feature diffusion range overlap between node A and node B is 0.32, which is greater than the threshold value 0.25. Therefore, in the propagation matrix, M[A, B]=1, indicating that there is a propagation link from A to B.

[0089] Based on propagation links, branch nodes are concatenated into propagation sequences and the temporal continuity of the nodes in the propagation sequence is checked. Starting from each starting node (a node with in-degree 0), a depth-first search algorithm is used to construct possible propagation sequences based on the link relationships in the propagation matrix M. For each constructed sequence, the timestamps of the nodes in it are checked to see if they meet the temporal continuity constraint, that is, the timestamp of the subsequent node in the sequence must be greater than the timestamp of the previous node.

[0090] For example, for the sequence [A, B, C, D], check the timestamp relationships: T(B) > T(A), T(C) > T(B), T(D) > T(C). If all relationships are satisfied, the sequence satisfies the temporal continuity constraint.

[0091] Propagation sequences that meet timing constraints and are free of loops are identified as propagation paths. The cumulative feature diffusion distance between adjacent nodes along the propagation path is calculated. For each propagation sequence that meets timing constraints, a check is performed to determine whether loops exist. Using a node marking method, visited nodes are marked during traversal. If a marked node is visited again, a loop exists and the sequence cannot be considered a valid propagation path.

[0092] For a determined propagation path, the feature diffusion cumulative distance D between adjacent nodes on the path is calculated. For an adjacent node pair (i, j) on the path, its feature diffusion cumulative distance D(i, j) takes into account the combined effects of spatial distance and time difference.

[0093] For example, for the propagation path [A, B, C], the characteristic diffusion cumulative distance D(A, B) from A to B and the characteristic diffusion cumulative distance D(B, C) from B to C are calculated, and then they are added together to obtain the total characteristic diffusion cumulative distance of the path D(A, B, C) = D(A, B) + D(B, C).

[0094] In the field of blockchain storage, the existing technology realizes content integrity verification by calculating the hash value of the content and storing it on the blockchain. At the same time, the digital content tracing technology uses timestamp and signature chain to realize content propagation tracking, and constructs the propagation path by recording the time point and operator of content change. In specific implementation, the existing technology usually records the hash value of content change at fixed time intervals, and uses a simple chain structure to record the content propagation process. In addition, the existing technology determines the relevance of content change by a single threshold, and constructs the evolution path of the content by sequential traversal.

[0095] The technical scheme proposed in this embodiment improves the existing technology in many aspects. In the storage mechanism, the timestamp and spatial location information are integrated into the storage process, and a feature diffusion intensity function containing exponential decay and Gaussian decay is used to describe the influence range of content change, realizing accurate control of tamper-proof verification. In the correlation analysis, the correlation degree of content change is evaluated by spatial grid sampling, and an overlap threshold is introduced to dynamically determine the content evolution relationship, and a propagation matrix reflecting global change characteristics is constructed. In the verification mechanism, the time sequence continuity constraint check is added to ensure the reliability of the storage, and the loop detection is used to avoid circular dependency, and the cumulative feature distance considering the time and space comprehensive influence is calculated. The accuracy of content tamper-proofing is improved, the description ability of change characteristics is enhanced, and the reliability of storage verification is improved.

[0096] Through actual verification, the scheme of this embodiment shows obvious advantages in many aspects. The content tamper-proofing model is more in line with the actual application requirements, and the verification accuracy is improved compared with the existing technology; the false positive rate of content change correlation determination is reduced; the identified content evolution path shows better time sequence consistency and logical continuity; the storage verification efficiency is improved, which can effectively support large-scale content tamper-proofing applications. These improvements enable the present scheme to more accurately describe the change characteristics and evolution process of digital content, and provide a more reliable tamper-proofing verification mechanism for the blockchain storage system.

[0097] As Figure 2The multi-dimension performance comprehensive comparison analysis table fully shows the differences between the technical solution and the other three traditional methods in various aspects. From the accuracy index, the technical solution has obvious advantages in all key indicators such as path recognition accuracy (98.3%), noise node filtering rate (97.9%), link integrity (96.8%), and timing accuracy (98.9%). In particular, the cumulative distance error is only 5.78, which is much lower than the traditional gradient diffusion (42.31), the pure spatial decay model (63.48), and the time window model (24.67), which fully shows that the technical solution can more accurately describe the propagation path of features between nodes. In terms of efficiency, the technical solution also performs outstandingly, with an average calculation time of only 47.5 ms and a memory occupation of only 8.3 MB, which is 65% and 70% less than the traditional gradient diffusion, respectively. More importantly, in terms of large-scale network scalability, the technical solution can achieve a high efficiency of 1250 nodes per second, which is 3.2 times that of the traditional gradient diffusion (385 nodes / second), reflecting the significant performance advantages brought by the time complexity of O(n log n) and the space complexity of O(n). The parallel processing efficiency reaches 4.8 times speedup, which is also significantly better than other methods. The robustness index further highlights the advantages of the technical solution. The noise tolerance reaches 0.55, the accuracy can still be maintained at 92.7% under data missing conditions, the abnormal value resistance is 94.2%, and the multi-path distinguishing ability reaches 95.7%, which shows that the technical solution can stably operate and maintain high precision in complex and noisy environments. In terms of application scene adaptability, the technical solution performs "excellent" in all scenes such as large-scale network, high-noise environment, long-chain propagation tracking, and real-time processing capability, showing wide applicability and stability. This is crucial for actual deployment, indicating that the technical solution can be reliably applied in various complex environments.

[0098] In the embodiment, by combining the time attenuation term and the space attenuation term, the feature diffusion intensity function is established, the feature diffusion range is accurately described, and the accuracy of propagation analysis is improved; the propagation matrix is established based on the overlap degree of the feature diffusion range, the propagation relationship can be automatically identified, and the dynamic propagation mode of different time and space scales can be adapted; the depth-first search method is used to construct the propagation sequence, and the timing continuity constraint and loop detection are used to ensure that the generated propagation path is reasonable and conforms to the actual propagation process; the cumulative distance of feature diffusion between adjacent nodes on the propagation path is calculated, and the time and space factors are considered comprehensively, to provide quantitative indicators for optimization and analysis of the propagation path.

[0099] In an optional implementation, a dynamic perception hash network is constructed based on the content association network, a graph neural network is used to perceive the state change of the associated content, and a dynamic hash value containing associated information is generated for each content node, which includes:

[0100] An initial state vector of a node is constructed based on a content feature of the node in a content-related network, and a connection strength feature of the node is calculated according to a connection edge weight between adjacent nodes; the content feature and the connection strength feature are combined to construct a node state vector; a connection edge feature is obtained by combining and encoding the node state vector of an adjacent node and the connection edge weight, and the connection edge feature reflects a connection strength change between node pairs;

[0101] A shortest path distance between nodes is calculated by using a spatial propagation function and is subjected to multi-level division, a node state vector in each level is subjected to weighted aggregation and cascaded fusion based on an exponential attenuation coefficient of a level sequence number and a normalized weight of a connection edge feature strength, a multi-scale neighborhood representation of the node is generated by combining scale self-adaptation and residual mapping, and a spatio-temporal fusion feature of the node is generated by fusing the multi-scale neighborhood representation and historical state information of the node by using a recursive state transition unit;

[0102] A hash feature is obtained by performing nonlinear transformation on the spatio-temporal fusion feature, and an initial hash value is obtained by binarizing and quantizing the hash feature; a propagation influence range of a node state change is calculated based on a node connection topology structure, the initial hash value is dynamically adjusted according to a state change of a connected node in the propagation influence range, and a dynamic hash value that can adaptively reflect a node state change is obtained.

[0103] In a specific embodiment, an initial state vector of a node is constructed based on a content feature of the node in a content-related network. Specifically, for each node in the content-related network, a multi-modal content feature of the node is extracted, such as text, image or video. For example, for text content, a BERT model is used to extract a 768-dimensional semantic feature vector; for image content, a ResNet50 is used to extract a 2048-dimensional visual feature vector. After dimension unification processing (such as mapping to 256 dimensions by a fully connected layer), the feature vectors are used as the initial state vector of the node.

[0104] A connection strength feature of the node is calculated according to a connection edge weight between adjacent nodes, and the connection edge weight can be determined by similarity, interaction frequency or co-occurrence probability of node content. For example, the text similarity of two content nodes is 0.85, and the edge weight is set to 0.85. For each node, the edge weights of all adjacent nodes are counted, and statistical quantities such as average value, maximum value, minimum value and standard deviation are calculated to form a 16-dimensional connection strength feature vector.

[0105] The content feature and the connection strength feature are combined to construct a node state vector. Specifically, the 256-dimensional content feature vector and the 16-dimensional connection strength feature vector are spliced, and then mapped to 128 dimensions by a fully connected layer to form a comprehensive state vector of the node.

[0106] The node state vectors and the connection edge weights of adjacent nodes are combined and encoded to obtain the associated edge features. For each pair of connected nodes i and j in the network, their state vectors are denoted as Vi and Vj, and the connection edge weight is Wij. By splicing operation [Vi, Vj, Wij] and passing through a multi-layer perception network, a 64-dimensional associated edge feature vector is generated, which reflects the change of the association strength between the node pair.

[0107] In the spatial propagation stage, the shortest path distance between nodes is calculated using a spatial propagation function and multi-level division is performed. An improved Dijkstra algorithm is used to calculate the shortest path distance between any two nodes in the network. Exemplarily, for a network containing 1000 nodes, the distance between nodes can be divided into 3 levels: nodes within 1 hop distance are the first level, nodes within 2-3 hop distances are the second level, and nodes within 4-6 hop distances are the third level.

[0108] Based on the exponential decay coefficient of the level number and the normalized weight of the associated edge feature strength, the node state vectors in each level are weighted, converged and cascaded. Specifically, for the kth level (k = 1, 2, 3), the decay coefficient is set to exp(-k), i.e. 0.368 for the first level, 0.135 for the second level, and 0.050 for the third level. For each node in each level, the normalized weight is calculated according to the associated edge feature strength, and the weight value ranges from 0 to 1, and the weight sum is 1. For example, the weights of three neighbor nodes of a node are 0.5, 0.3 and 0.2 respectively. The node state vectors in each level are weighted and summed according to the weights to obtain the converged features of each level, and then the features of the three levels are spliced to form a 384-dimensional (128x3) multi-level fusion feature.

[0109] The multi-level fusion feature is processed by convolution layers with three different kernel sizes (such as 1x1, 3x3, 5x5) to capture neighborhood information at different scales. At the same time, residual connection is introduced to add the original feature and the convolved feature, enhancing the information flow. Finally, a 192-dimensional multi-scale neighborhood representation is generated.

[0110] The multi-scale neighborhood representation and the historical state information of the node are fused by a recursive state transition unit to generate the spatio-temporal fusion feature of the node. A gated recurrent unit (GRU) is used as the recursive state transition unit, the input is the multi-scale neighborhood representation (192-dimensional) at the current time and the hidden state (128-dimensional) at the previous time, and the output is the updated hidden state (128-dimensional) as the spatio-temporal fusion feature of the node. For example, at three time points t = 1, 2, 3, the node state evolves into different feature vectors, reflecting the time series change characteristics.

[0111] A nonlinear transformation is performed on the spatiotemporal fusion features to obtain hash features. A two-layer fully connected network maps the 128-dimensional spatiotemporal fusion features into a 64-dimensional hash feature space. The first layer uses the ReLU activation function, while the second layer does not, to preserve the distribution characteristics of the features.

[0112] The hash feature is binarized and quantized to obtain an initial hash value. For each dimension of the 64-dimensional hash feature, values ​​greater than 0 are quantized to 1, and values ​​less than or equal to 0 are quantized to 0. This generates a 64-bit binary hash code as the initial hash value. For example, the initial hash value obtained after quantization of the hash feature vector of a node is "1010...0101".

[0113] The propagation range of node state changes is calculated based on the node-association topology. The impact range of state changes is determined by analyzing network connectivity and node centrality. For example, for nodes with high centrality (e.g., degree centrality greater than twice the network average), the impact range of their state changes may extend to all nodes within a two-hop distance. For edge nodes, however, the impact range may be limited to directly connected neighboring nodes.

[0114] The initial hash value is dynamically adjusted based on the state changes of associated nodes within the propagation influence range, resulting in a dynamic hash value that can adaptively reflect node state changes. Specifically, the magnitude of the state change of all nodes within the influence range is calculated, and the hash bits that need to be flipped are determined based on the magnitude of the change and the strength of the association between the nodes. For example, if a significant change in the state of a strongly associated node is detected (the magnitude of the change exceeds a threshold of 0.3), 2-3 bits in the initial hash value may be flipped, generating a new dynamic hash value "1110...0101" to reflect the dynamic changes in the network state.

[0115] In an optional embodiment, the shortest path distance between nodes is calculated using a spatial propagation function and multi-level division is performed. Based on the exponential decay coefficient of the level number and the normalized weight of the associated edge feature strength, the node state vectors in each level are weighted aggregated and cascaded. The multi-scale neighborhood representation of the node is generated by combining scale adaptation and residual mapping, including:

[0116] Calculate the shortest path distance based on the connection path between nodes using the spatial propagation function, divide the neighborhood nodes of the central node into levels according to the shortest path distance, and divide the neighborhood nodes within different distance ranges from the central node into different levels according to the preset range intervals, and determine the level sequence number;

[0117] The spatial propagation function calculates a distance attenuation coefficient corresponding to each level according to a level sequence number, and makes the distance attenuation coefficient decrease with the increase of the level through an exponential attenuation function; and for each target node in each level, an association strength weight is calculated based on an association edge feature strength between the target node and a center node, and the association edge feature strength is normalized in the level to obtain the association strength weight;

[0118] The spatial propagation function multiplies a state vector of each node in each level by a distance attenuation coefficient corresponding to the level, and multiplies the state vector by an association strength weight corresponding to the node and sums in the level to generate a node feature of the level; and the node features of all levels are concatenated in a level sequence to generate an initial multi-scale neighborhood representation through a nonlinear transformation;

[0119] A scale adaptation weight is calculated for the node feature of each level, and the scale adaptation weight is used to weight and fuse the node features of each level; and a residual mapping is performed on the initial multi-scale neighborhood representation, and a mapping result is added to the weighted and fused features to generate an optimized multi-scale neighborhood representation.

[0120] In a specific embodiment, a spatial propagation function is used to calculate the shortest path distance between nodes and perform multi-level division. In a network structure, for a center node v0 and any node vi in the neighborhood of the center node, the shortest path distance d(v0, vi) between the two nodes is calculated by a breadth-first search or Dijkstra algorithm. Taking a social network as an example, the shortest path distance between user A and user B is 2, indicating that an intermediate user is needed to connect A to B.

[0121] Based on the calculated shortest path distance, the neighborhood nodes are divided into K levels. Specifically, K is set to 3, and the distance range intervals are [1, 1], [2, 2], and [3, 4]. The first level contains nodes directly connected to the center node (distance 1), the second level contains nodes with a distance of 2, and the third level contains nodes with a distance of 3 or 4. For example, in the above social network, the first level of user A contains its direct friends, the second level contains friends of friends, and the third level contains indirectly associated users with a distance of 3 or 4.

[0122] Next, a distance attenuation coefficient is calculated based on the level sequence number. For the kth level, the distance attenuation coefficient γk is calculated by an exponential attenuation function: γk=α (k-1) , where α is an attenuation base with a value range of (0, 1). In actual application, α can be set to 0.5, and the attenuation coefficients of the three levels are γ1=1, γ2=0.5, and γ3=0.25, respectively, which reflects the characteristic of information transmission attenuation with the increase of the level.

[0123] Meanwhile, for each node in each level, the correlation strength weight is calculated based on the correlation edge feature strength between the node and the center node. The correlation edge feature strength can be the weight of the edge, the interaction frequency, etc. Taking a social network as an example, the interaction times of user A with users B, C and D in the first level of user A are 10, 5 and 2 respectively, and the normalized correlation strength weights are 0.59, 0.29 and 0.12 respectively.

[0124] In the feature aggregation stage, the spatial propagation function weights and converges the node features in each level. Assuming that the state vector dimension of the node is 4, the first level of the center node v0 contains nodes v1, v2 and v3, and the state vectors of the nodes are [0.1, 0.2, 0.3, 0.4], [0.2, 0.3, 0.4, 0.5] and [0.3, 0.4, 0.5, 0.6] respectively, the correlation strength weights are 0.5, 0.3 and 0.2 respectively, and the distance decay coefficient γ1 = 1. Then the node feature of the first level is calculated as: [0.1, 0.2, 0.3, 0.4] * 0.5 * 1 + [0.2, 0.3, 0.4, 0.5] * 0.3 * 1 + [0.3, 0.4, 0.5, 0.6] * 0.2 * 1 = [0.16, 0.26, 0.36, 0.46].

[0125] Similarly, the node features of the second level and the third level are calculated, and are denoted as h2 and h3 respectively. The features of all levels are concatenated in order to form a vector [h1, h2, h3], and an initial multi-scale neighborhood representation h_init is generated through a nonlinear transformation (such as using a ReLU activation function).

[0126] In order to further optimize the feature representation, the application introduces a scale adaptive mechanism and a residual mapping. First, the scale adaptive weight is calculated for the node features of each level. Through the attention mechanism, a weight βk is assigned to each level feature hk, and the sum of all βk is 1. For example, the weights of the three levels can be β1 = 0.6, β2 = 0.3 and β3 = 0.1, indicating that the closer the level information, the greater the contribution. k For example, the weights of the three levels can be β1 = 0.6, β2 = 0.3 and β3 = 0.1, indicating that the closer the level information, the greater the contribution.

[0127] The scale adaptive weight is used to weight and fuse the hierarchical features: h_fusion = β1 × h1 + β2 × h2 + β3 × h3. In the above example, if h1 = [0.16, 0.26, 0.36, 0.46], h2 = [0.08, 0.13, 0.18, 0.23], and h3 = [0.04, 0.06, 0.09, 0.11], then h_fusion = [0.6 × 0.16 + 0.3 × 0.08 + 0.1 × 0.04, 0.6 × 0.26 + 0.3 × 0.13 + 0.1 × 0.06, 0.6 × 0.36 + 0.3 × 0.18 + 0.1 × 0.09, 0.6 × 0.46 + 0.3 × 0.23 + 0.1 × 0.11] = [0.128, 0.208, 0.288, 0.368].

[0128] The residual mapping is performed on the initial multi-scale neighborhood representation h_init, which is mapped to a vector h_res of the same dimension as h_fusion through a fully connected layer, and then the two are added to obtain the optimized multi-scale neighborhood representation h_final = h_fusion + h_res.

[0129] In the field of blockchain notarization, the existing digital content node feature extraction technology mainly adopts the message passing mechanism of graph neural networks. Traditional methods obtain local structural information of nodes through simple neighborhood aggregation operations and use fixed aggregation weights for feature fusion. This method has limitations when dealing with multi-scale features of digital content nodes, and cannot effectively capture the dynamic association relationships between nodes in different distance ranges. In specific implementation, existing technologies usually use single-layer neighborhood aggregation or use pre-defined fixed weights for multi-layer feature combination, which leads to a lack of adaptability of the feature extraction process to the strength of the relationship between nodes and cannot reflect the importance of different scale features.

[0130] The present embodiment significantly improves the expression ability of digital content node features by introducing a multi-level feature extraction mechanism based on a spatial propagation function. First, in spatial structure modeling, multi-level division of nodes is achieved through shortest path distance calculation, and an exponentially decaying distance attenuation coefficient is introduced, so that the feature propagation process can accurately reflect the spatial relationship between nodes. Secondly, in association strength measurement, more accurate node association weight calculation is achieved through intra-level normalization of associated edge features. In the feature fusion stage, a scale adaptive mechanism is innovatively designed, which dynamically learns the importance weights of different levels to achieve adaptive feature fusion. At the same time, a residual mapping structure is introduced to further enhance the non-linear ability of feature expression. These improvements aim to improve the accuracy and adaptability of feature extraction and enhance the perception ability of node representation to multi-scale structural information.

[0131] Experimental verification demonstrates that the improved technical solution achieves significant improvements across multiple performance metrics. The enhanced expressiveness of node features improves accuracy in content similarity judgment tasks. The adaptive fusion mechanism of multi-scale features enables the model to better adapt to node relationships in diverse scenarios, enhancing generalization performance. The introduced residual structure effectively alleviates optimization issues in deep networks, accelerating model convergence. These improvements enable this solution to more accurately characterize the structural characteristics of digital content nodes, providing a more reliable feature representation foundation for subsequent anti-tampering verification.

[0132] like Figure 3 As shown in the figure, the performance comparison of the multi-scale neighborhood representation in different performance dimensions before and after optimization is shown through a parallel coordinate graph. From the data in the figure, it can be observed that after the weighted fusion of residual mapping and scale-adaptive weights, this technical solution has significantly improved in six key indicators. In particular, in terms of feature distinguishability, the optimized multi-scale neighborhood representation reached 0.87, while the initial representation was only 0.71, an improvement of 22.5%. In terms of information retention rate, it reached 0.91 after optimization, which is 16.7% higher than the initial 0.78, indicating that the residual mapping effectively retains the original feature information. In terms of structural sensitivity index, the performance after optimization is 0.89, while GCN is only 0.67 and GraphSAGE is 0.72, which fully proves that this technical solution can more effectively capture graph structure information. It is worth noting that in terms of computational efficiency, although this technical solution is slightly lower than the initial representation (from 0.83 to 0.79), it is still better than GAT's 0.71. Overall, this technical solution achieved an average score of 0.87 across all evaluation metrics, significantly higher than other comparison methods, including GCN (0.70), GraphSAGE (0.74), and GAT (0.76). These data fully demonstrate that the multi-scale neighborhood representation optimization method proposed in this technical content can significantly improve the expressiveness and robustness of features while maintaining computational efficiency.

[0133] In this embodiment, by hierarchical division of the neighborhood nodes of the central node and multi-scale feature fusion, comprehensive extraction of information at different levels from near to far is achieved, thereby improving the expressive power of the neighborhood representation; an exponential decay function is used to assign different distance attenuation coefficients to nodes at each level, so that the contribution of nodes farther away from the central node to the feature representation gradually weakens, thereby better reflecting the attenuation law of spatial influence; by normalizing the strength of the associated edge features and calculating the weights, the association strength between each neighborhood node and the central node can be effectively reflected, thereby enhancing the expression of key connection information; the scale-adaptive weights are used to perform weighted fusion of node features at each level, and are optimized through residual mapping, thereby improving the robustness and discrimination ability of the overall neighborhood representation.

[0134] In an alternative embodiment, the hierarchical verification groups are formed based on the node coverage degree calculated from the evolutionary phylogenetic tree in the blockchain verification pool, the dynamic hash value is verified through the progressive consensus mechanism, and the verification result is passed layer by layer to generate a record of evidence containing the dynamic hash value, the evolutionary phylogenetic tree information and the timestamp, including:

[0135] The distance value between nodes is calculated based on the evolutionary phylogenetic tree in the blockchain verification pool, and the distance value between nodes is dynamically adjusted and weighted calculated by combining the node reputation and the adaptive decay factor to obtain the local coverage degree of the node; the global coverage degree of the node is calculated based on the ratio of the local coverage degree to the sum of the local coverage degrees of all nodes in the verification pool;

[0136] The reference threshold value and the exponential operation result of the hierarchical decay rate are used as the hierarchical threshold value, the nodes are layered according to the hierarchical threshold value, and the nodes with a global coverage degree greater than the corresponding hierarchical threshold value are divided into the verification group of the corresponding level;

[0137] The nodes in each verification group verify the dynamic hash value and generate a verification signature, and the ratio of the number of verification signatures in the verification group to the total number of nodes in the verification group is calculated as the consensus degree of the verification group. When the consensus degree of the verification group is greater than the preset consensus threshold, the verification result is passed to the upper layer verification group;

[0138] The dynamic hash value, the evolutionary phylogenetic tree information, the timestamp and the verification signature of each verification group are integrated to generate a blockchain record of evidence.

[0139] In a specific embodiment, an evolutionary phylogenetic tree is constructed in the blockchain verification pool, which records the kinship and evolution history between nodes. When each node joins the verification pool, a unique identifier is calculated based on its characteristic information (such as public key, IP address, computing power, etc.), and the most similar node is found as its parent node in the evolutionary phylogenetic tree to determine its position in the tree.

[0140] When calculating the distance value between nodes based on the evolutionary phylogenetic tree, the path distance calculation method in the tree structure is adopted. For example, if the nearest common ancestor of node A and node B in the tree is node C, then the distance between A and B is the distance between A and C plus the distance between B and C. Specifically, if the distance between A and C is 3 (i.e. 3 edges are needed), and the distance between B and C is 2, then the distance between A and B is 5.

[0141] The inter-node distance value needs to be dynamically adjusted in combination with node reputation and an adaptive decay factor. The node reputation is evaluated based on historical behavior, including verification accuracy, online duration, response speed, etc. For example, if the reputation of node A is 0.85 (full score is 1), its distance value with other nodes will be multiplied by the inverse of 0.85, about 1.18, to increase the distance accordingly. The adaptive decay factor is adjusted in real time according to network congestion, verification task complexity, etc. When the network congestion is 60%, the decay factor may be 0.75, further adjusting the distance value.

[0142] The node local coverage is calculated by weighting, representing the coverage ability of the node to the surrounding area. When calculating, the adjusted distance value of the node with all other nodes is taken as the inverse, multiplied by the respective weight and summed. For example, the adjusted distance of node A with three nodes is 6, 8, 10 respectively, and the corresponding weight is 0.5, 0.3, 0.2, then the local coverage of A is 0.5 x (1 / 6) + 0.3 x (1 / 8) + 0.2 x (1 / 10) = 0.1125.

[0143] The global coverage is calculated by dividing the local coverage of the node by the sum of the local coverages of all nodes in the verification pool. Assuming there are 100 nodes in the verification pool, the sum of the local coverages of all nodes is 15, then the global coverage of node A is 0.1125 ÷ 15 = 0.0075, representing its relative importance in the entire verification network.

[0144] The formation of the hierarchical verification group is based on the hierarchical threshold, which is determined by the benchmark threshold and the exponential operation result of the hierarchical decay rate. Specifically, if the benchmark threshold is 0.01 and the hierarchical decay rate is 0.8, the threshold of the first layer is 0.01, the second layer is 0.01 x 0.8 = 0.008, the third layer is 0.01 x 0.8 2 = 0.0064, and so on. Nodes are assigned to the verification group of the corresponding layer according to the comparison result of their global coverage and the corresponding hierarchical threshold. For example, the global coverage of node A is 0.0075, which is greater than the second layer threshold 0.008 but less than the first layer threshold 0.01, so it is assigned to the second layer verification group.

[0145] When verifying the dynamic hash value, a progressive consensus mechanism is adopted. First, the nodes of the bottommost layer verification group verify the dynamic hash value, and the verification method includes checking the format, length, whether it meets the output characteristics of a specific algorithm, etc. Each node generates a verification signature after completing the verification, which contains node ID, verification result (pass / fail), timestamp, etc. and is encrypted using the node private key.

[0146] The consensus of a validation group is calculated as the ratio of the number of validation signatures in the group to the total number of nodes in the validation group. For example, if a third-tier validation group has 20 nodes and 18 of them have generated validation signatures, the consensus of the group is 18 ÷ 20 = 0.9. If the preset consensus threshold is 0.75, then 0.9 > 0.75, and the validation result can be passed to the next higher-tier validation group.

[0147] After receiving the verification results from the lower layers, the upper-level verification group will conduct a secondary verification, verifying not only the dynamic hash value itself but also the validity of the signatures of the lower-level verification groups and whether the consensus level has been met. Verification groups at each layer follow the same mechanism for verification and result transmission until the highest-level verification group completes the verification.

[0148] The dynamic hash value of each layer of verification, the evolutionary pedigree tree information (including the position of the nodes involved in the verification in the tree and the relationship between the nodes), the timestamp with millisecond accuracy, and the verification signatures of each verification group are integrated to generate a blockchain evidence record. This record uses a specific format, such as JSON structure, and contains a hash value field, tree information field, timestamp field, and signature array field. It is signed and confirmed by the nodes of the highest-level verification group before being written to the blockchain.

[0149] In this embodiment, node coverage calculation and hierarchical verification group formation based on the evolutionary pedigree tree are implemented, the progressive consensus mechanism is used to improve verification efficiency and reliability, and a blockchain evidence record containing complete verification information is generated, providing a new solution for blockchain data verification.

[0150] In an optional embodiment, the distance value between nodes is calculated based on the evolutionary pedigree tree in the blockchain verification pool, and the distance value between nodes is dynamically adjusted and weighted in combination with the node reputation and the adaptive attenuation factor to obtain the node local coverage including:

[0151] Calculate the node distance matrix based on the evolutionary lineage tree in the blockchain verification pool and determine the node distance value; calculate the node reputation based on the historical verification performance of the nodes in the blockchain verification pool, and the node reputation includes the node verification accuracy, node response timeliness and node verification consistency;

[0152] Constructing an adaptive attenuation factor, wherein the adaptive attenuation factor is composed of a preset basic attenuation rate and network dynamic parameters, wherein the network dynamic parameters include the current verification load of the blockchain verification pool, the network congestion of the blockchain verification pool, and the node activity of the blockchain verification pool; and dynamically adjusting the inter-node distance value according to the adaptive attenuation factor;

[0153] A node weight coefficient is set based on an influence degree of the node reputation on neighborhood nodes in the blockchain validation pool, and a node evolution correlation strength is obtained by weighting calculation of the node weight coefficient and a dynamically adjusted inter-node distance value.

[0154] In a specific embodiment, an inter-node distance matrix is calculated based on an evolution pedigree tree in the blockchain validation pool to determine an inter-node distance value. Evolution pedigree information of nodes is extracted from the blockchain validation pool to construct an evolution pedigree tree. Assuming that there are five nodes A, B, C, D, and E in the validation pool, the evolution distance between the nodes can be obtained by analyzing the common ancestors and differentiation times of the nodes. For example, the distance value between node A and node B is 0.3, the distance value between node A and node C is 0.5, the distance value between node A and node D is 0.7, and the distance value between node A and node E is 0.9. These distance values form a distance vector of node A, and the distance vectors of other nodes can be obtained in the same way to form a complete inter-node distance matrix.

[0155] The node reputation is calculated according to the historical validation performance of the nodes in the blockchain validation pool, including three aspects: node validation accuracy, node response timeliness, and node validation consistency. In actual application, the system will record the historical validation records of each node. For example, for node A, 95 out of the last 100 validation results are correct, so the validation accuracy is 0.95; the average response time is 200 milliseconds, which is 0.67 relative to the average response time of 300 milliseconds of the validation pool; in the validation result consistency comparison with other nodes, 90% of the validation results are consistent with most nodes, so the validation consistency is 0.9. Combining these three indicators, the comprehensive reputation of node A can be calculated as 0.84.

[0156] An adaptive decay factor is constructed, which is composed of a preset basic decay rate and network dynamic parameters. In this embodiment, the basic decay rate is set to 0.05 as the initial decay reference. The network dynamic parameters include the current validation load of the blockchain validation pool, network congestion, and node activity. Assuming that the validation load of the current validation pool is 75% (relative to the maximum processing capacity), the network congestion is 0.3 (network bandwidth occupancy rate), and the node activity is 0.8 (the proportion of active nodes to total nodes). According to these parameters, the system calculates a dynamic adjustment value of 0.04 under the current network state. Combining the basic decay rate and the dynamic adjustment value, the adaptive decay factor is 0.09.

[0157] The distance between nodes is dynamically adjusted based on an adaptive attenuation factor. For example, if the distance between nodes A and B is 0.3, applying an adaptive attenuation factor of 0.09 results in a distance of 0.273. This adjustment reflects the impact of the current network state on inter-node relationships. When the network is highly loaded or congested, the effective distance between nodes is shortened, fostering closer collaboration.

[0158] Node weights are set based on the influence of a node's reputation on its neighboring nodes in the blockchain verification pool. In this example, node A has a reputation of 0.84, and based on the preset weight mapping, its weight is determined to be 0.75. For node B, its reputation is 0.92, and its corresponding weight is 0.85. These weights reflect the importance and reliability of a node in the verification network.

[0159] The node weight coefficient and the dynamically adjusted inter-node distance are weighted together to calculate the evolutionary correlation strength between nodes. For example, consider nodes A and B. Node A's weight coefficient is 0.75, and Node B's weight coefficient is 0.85. The adjusted distance is 0.273, resulting in an evolutionary correlation strength of 0.218 between the two nodes. This value reflects the actual degree of correlation between the two nodes after factoring in their reputations.

[0160] The local coverage of a node is obtained by summing the evolutionary correlation strengths between nodes. Taking node A as an example, the sum of its evolutionary correlation strengths with all other nodes (B, C, D, and E) is calculated. Assuming that the evolutionary correlation strengths of node A with node B are 0.218, with node C is 0.195, with node D is 0.168, and with node E is 0.142, the local coverage of node A is 0.723. This value represents the influence and coverage of node A in the entire verification pool network.

[0161] In practice, the inter-node distance matrix, node reputation, and adaptive decay factor are regularly updated to reflect changes in network status and node performance. For example, after verifying every 100 blocks, the system recalculates all parameters and updates the node local coverage. This dynamic adjustment mechanism ensures that the validation pool can adapt to changes in the network environment and optimizes the efficiency of inter-node collaboration.

[0162] The node local coverage calculated by the above method can be used in various scenarios such as verification task allocation, consensus mechanism optimization, and malicious node detection to improve the overall performance and security of the blockchain verification pool.

[0163] The existing verification node management technology mainly adopts a fixed weight trust mechanism and a simple distance measurement method. The traditional scheme usually evaluates the node credibility based on the verification times or survival time of the node, and calculates the distance between nodes using a static network topology. This method has obvious shortcomings when dealing with large-scale dynamic verification networks: it cannot accurately reflect the dynamic changes of node verification behavior, it is difficult to adapt to real-time fluctuations in network load, and the evaluation of the correlation strength between nodes is too simple. Specifically, existing technologies often use a single-dimensional node evaluation indicator, ignoring important features such as node response timeliness and verification consistency, and lack adaptive adjustment mechanisms for network dynamic characteristics.

[0164] In terms of node reputation quantification, the embodiment introduces a comprehensive evaluation mechanism with three dimensions of verification accuracy, response timeliness, and verification consistency, making node evaluation more comprehensive and objective. In terms of network dynamic adaptability, by constructing adaptive decay factors including verification load, network congestion, and node activity, dynamic adjustment of the distance between nodes is achieved. Meanwhile, in terms of node correlation evaluation, the distance calculation method based on the evolutionary pedigree tree provides a more accurate measurement of node relationships, combined with the weighted calculation mechanism of node weight coefficients, accurate evaluation of the influence range of nodes is achieved. These improvements aim to improve the dynamic adaptability of the verification pool and enhance the accuracy and reliability of node management.

[0165] Experimental results show that the scheme of the embodiment achieves significant improvement in multiple key indicators. In terms of verification task allocation efficiency, the multi-dimensional reputation evaluation mechanism improves the verification success rate; in terms of network adaptability, the adaptive decay mechanism improves the stability of the system under load fluctuations; in terms of node collaboration efficiency, the node coverage calculation based on evolutionary correlation strength improves the overall processing efficiency of the verification pool. The scheme of the embodiment can better support large-scale blockchain verification scenarios, significantly improving the reliability and efficiency of the verification pool. Especially under high load conditions, the system shows stronger robustness and adaptability, providing more reliable technical support for digital content tamper-proof storage.

[0166] As shown in Figure 4 and Figure 5 , the present technical solution is compared with the traditional weighted distance algorithm and the static pedigree tree algorithm in five key performance indicators. From the chart data, it can be clearly seen that the scheme of the embodiment shows obvious advantages in all evaluation dimensions. In terms of verification accuracy, the technical solution achieves a high accuracy of 97.8%, which is 8.5 percentage points higher than the traditional weighted distance algorithm and 19.3 percentage points higher than the static pedigree tree algorithm. This significant improvement is mainly due to the combination of evolutionary pedigree tree calculation and adaptive decay factor, which enables the system to more accurately identify and process verification requests.

[0167] In terms of node coverage, this technical solution reached 86.7%, which is an increase of 12.4% and 23.9% respectively compared to the 74.3% of the traditional weighted distance algorithm and 62.8% of the static pedigree tree algorithm. This is the most significant performance improvement. This shows that the setting of weight coefficients based on node credibility can effectively expand the influence range of highly trusted nodes and optimize the resource utilization efficiency of the entire verification pool. Although the response timeliness index seems to be reduced (the value has increased), this actually reflects the system's moderate trade-off in time while maintaining high accuracy and consistency. Judging from the response time of 8.2ms, it still maintains an extremely high level of efficiency. The verification consistency reached 92.5%, which is also significantly higher than the other two algorithms, proving that this technical solution can maintain a higher level of consensus in a distributed environment.

[0168] In this embodiment, by combining the distance matrix calculated based on the evolutionary pedigree tree of the nodes in the blockchain verification pool and introducing an adaptive attenuation factor (taking into account dynamic parameters such as the current verification load, network congestion, and node activity), real-time and dynamic adjustment of the distance values ​​between nodes is achieved, thereby more accurately reflecting the actual evolutionary relationship between nodes; the node's historical verification performance (verification accuracy, response timeliness, and verification consistency) is used to calculate the node's reputation, and the weight coefficient of the influence of neighboring nodes is set accordingly, effectively distinguishing and giving priority to high-reputation nodes, thereby improving the overall security and reliability of the system; the node reputation and the dynamically adjusted distance value are weighted and calculated to obtain the evolutionary correlation strength between nodes, and the local coverage of the nodes is further quantified by accumulation, providing an accurate and quantitative basis for subsequent network analysis or decision-making.

[0169] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0170] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A distributed multi-source digital content tamper-proof hash evidence storage method based on blockchain, characterized by: include: receiving multi-source digital content, and converting the multi-source digital content into standardized data; Constructing an evolutionary pedigree tree of the standardized data, implanting detectors at branch nodes, capturing change features and tracing propagation paths, calculating correlation strength values ​​between data, and constructing a content association network; Building a dynamic perception hash network based on the content association network, using a graph neural network to perceive the state changes of associated content, and generating a dynamic hash value containing associated information for each content node; In the blockchain verification pool, a hierarchical verification group is formed based on the node coverage calculated based on the evolutionary pedigree tree. The dynamic hash value is verified through a progressive consensus mechanism. The verification results are passed layer by layer and a certificate record containing the dynamic hash value, evolutionary pedigree tree information and timestamp is generated. When receiving a query request, it obtains the evidence record and reconstructs the evolutionary lineage tree, calculates the hash value for verification, and returns the verification result.

2. The method according to claim 1, characterized in that Constructing an evolutionary pedigree tree of the standardized data, implanting detectors at branch nodes, capturing change features and tracing propagation paths, calculating correlation strength values ​​between data, and constructing a content association network includes: Taking the standardized data of multi-source digital content as the root node, constructing an evolutionary pedigree tree based on iterative versions of the standardized data, and constructing branch nodes by comparing the feature differences and time intervals of the standardized data in different iterative versions; A detector is implanted in the branch node, and the detector monitors and records in real time the characteristic change amount and propagation time interval of the standardized data when it passes through the corresponding branch node, and generates a change characteristic of the branch node; Constructing a feature diffusion intensity function based on the change feature, calculating the feature diffusion range of each branch node, determining a propagation path according to the degree of overlap of the feature diffusion range, and calculating the feature diffusion cumulative distance between adjacent nodes on the propagation path; The feature diffusion cumulative distance is used as the association strength value between branch nodes, and the association strength value decays exponentially with the increase of the feature diffusion cumulative distance; The branch nodes are used as network nodes, and the association strength values ​​higher than a preset strength threshold are used as weights of connection edges to construct a content association network of the standardized data.

3. The method according to claim 2, characterized in that Constructing a feature diffusion intensity function based on the change feature, calculating the feature diffusion range of each branch node, determining the propagation path according to the degree of overlap of the feature diffusion range, and calculating the feature diffusion cumulative distance between adjacent nodes on the propagation path includes: Based on the timestamp of the changed feature and the spatial position of the node, a feature diffusion intensity function is established, wherein the feature diffusion intensity function includes a time attenuation term and a spatial attenuation term to generate the feature diffusion range of each branch node; A propagation matrix is ​​established based on the overlapping area of ​​the feature diffusion range between branch nodes. When the overlap of the feature diffusion range of two branch nodes exceeds the preset diffusion threshold, the propagation link between the nodes is recorded at the corresponding position in the propagation matrix. Connecting branch nodes in series into a propagation sequence based on the propagation link, and detecting the temporal continuity of nodes in the propagation sequence; A propagation sequence that meets the timing constraint and has no loop is determined as a propagation path, and the characteristic diffusion cumulative distance between adjacent nodes on the propagation path is calculated.

4. The method according to claim 1, wherein Building a dynamic perception hash network based on the content association network, using a graph neural network to perceive the state changes of associated content, and generating a dynamic hash value containing association information for each content node includes: Constructing a node initial state vector based on content features of nodes in a content association network, and calculating node association strength features based on edge weights between adjacent nodes; combining the content features and the association strength features to construct a node state vector; and combining and encoding the node state vectors and edge weights of adjacent nodes to obtain an association edge feature, wherein the association edge feature reflects changes in association strength between node pairs; The shortest path distance between nodes is calculated using a spatial propagation function and multi-level division is performed. Based on the exponential decay coefficient of the level number and the normalized weight of the associated edge feature strength, the node state vectors in each level are weighted and aggregated and cascaded. Scale adaptation and residual mapping are combined to generate a multi-scale neighborhood representation of the node. The recursive state transfer unit is used to fuse the multi-scale neighborhood representation with the historical state information of the node to generate the spatiotemporal fusion feature of the node. The spatiotemporal fusion feature is nonlinearly transformed to obtain a hash feature, and the hash feature is binarized and quantized to obtain an initial hash value; the propagation influence range of the node state change is calculated based on the node association topology structure, and the initial hash value is dynamically adjusted according to the state change of the associated nodes within the propagation influence range to obtain a dynamic hash value that can adaptively reflect the node state change.

5. The method according to claim 4, characterized in that The shortest path distance between nodes is calculated using the spatial propagation function and multi-level division is performed. Based on the exponential decay coefficient of the level number and the normalized weight of the associated edge feature strength, the node state vectors in each level are weighted aggregated and cascaded. The multi-scale neighborhood representation of the node is generated by combining scale adaptation and residual mapping, including: Calculate the shortest path distance based on the connection path between nodes using the spatial propagation function, divide the neighborhood nodes of the central node into levels according to the shortest path distance, and divide the neighborhood nodes within different distance ranges from the central node into different levels according to the preset range intervals, and determine the level sequence number; The spatial propagation function calculates the distance attenuation coefficient corresponding to each level according to the level sequence number, and uses an exponential decay function to make the distance attenuation coefficient decrease as the level increases; and for the target node in each level, the association strength weight is calculated based on the association edge feature strength between the target node and the central node, and the association edge feature strength is normalized within the layer to obtain the association strength weight; The spatial propagation function multiplies the state vector of the node in each level by the distance decay coefficient corresponding to the level, and multiplies it by the association strength weight corresponding to the node and sums them within the level to generate the node features of the level; the node features of all levels are cascaded in the order of the levels, and the initial multi-scale neighborhood representation is generated through nonlinear transformation; Scale adaptation weights are calculated for node features at each level, and the node features at each level are weightedly fused using the scale adaptation weights. Residual mapping is performed on the initial multi-scale neighborhood representation, and the mapping result is added to the weighted fused features to generate an optimized multi-scale neighborhood representation.

6. The method according to claim 1, characterized in that In the blockchain verification pool, a hierarchical verification group is formed based on the node coverage calculated based on the evolutionary pedigree tree. The dynamic hash value is verified through a progressive consensus mechanism. The verification results are passed layer by layer and a record containing the dynamic hash value, evolutionary pedigree tree information and timestamp is generated. The records include: The distance between nodes is calculated based on the evolutionary lineage tree in the blockchain verification pool. The distance between nodes is dynamically adjusted and weighted based on the node reputation and adaptive attenuation factor to obtain the local coverage of the node. The global coverage of the node is calculated based on the ratio of the local coverage to the sum of the local coverage of all nodes in the verification pool. The result of the exponential operation of the baseline threshold and the level decay rate is used as the level threshold, and the nodes are layered according to the level threshold. The nodes with global coverage greater than the corresponding level threshold are divided into the verification group of the corresponding level; Nodes in each verification group verify the dynamic hash value and generate a verification signature. The ratio of the number of verification signatures in the verification group to the total number of nodes in the verification group is calculated as the verification group consensus. When the verification group consensus is greater than the preset consensus threshold, the verification result is transmitted to the upper-level verification group. The dynamic hash value of the consensus verification of each layer of verification group, the evolutionary pedigree tree information, the timestamp and the verification signature of each verification group are integrated to generate a blockchain evidence record.

7. The method according to claim 6, characterized in that The distance between nodes is calculated based on the evolutionary lineage tree in the blockchain verification pool. The distance between nodes is dynamically adjusted and weighted based on the node reputation and adaptive attenuation factor. The local coverage of nodes is obtained, including: Calculate the node distance matrix based on the evolutionary lineage tree in the blockchain verification pool and determine the node distance value; calculate the node reputation based on the historical verification performance of the nodes in the blockchain verification pool, and the node reputation includes the node verification accuracy, node response timeliness and node verification consistency; Constructing an adaptive attenuation factor, wherein the adaptive attenuation factor is composed of a preset basic attenuation rate and network dynamic parameters, wherein the network dynamic parameters include the current verification load of the blockchain verification pool, the network congestion of the blockchain verification pool, and the node activity of the blockchain verification pool; and dynamically adjusting the inter-node distance value according to the adaptive attenuation factor; A node weight coefficient is set based on the degree of influence of the node reputation on the neighboring nodes of the node in the blockchain verification pool, and the node weight coefficient is weightedly calculated with the dynamically adjusted inter-node distance value to obtain the inter-node evolutionary association strength; the inter-node evolutionary association strength is accumulated to obtain the node local coverage.

Images