Digital cultural product right confirmation and data sharing method and system based on block chain

Abstract

The invention provides a digital culture product right confirmation and data sharing method and system based on a block chain, and relates to the technical field of digital culture, and the method comprises the steps: generating a comprehensive trust score through a self-adaptive trust evaluation model by fusing timeliness, integrity and consistency trust scores, and constructing a self-evolution data trust network and an encrypted data synchronization channel. After smart contract verification, the access authority of the data requester is determined according to the trust score, access features are extracted through a bidirectional long-short term memory network and a multi-head attention mechanism, a proof chain node is constructed, a mapping relation with an authority confirmation certificate is established by using a graph neural network, and the access contribution degree is calculated through homomorphic encryption and federated learning; and finally, performing income distribution according to a sectional income distribution curve. According to the invention, the problems of difficult right confirmation and low data sharing credibility of digital culture products are effectively solved, and safe and controllable data sharing and income distribution are realized.

Description

Technical Field

[0001] The present invention relates to digital culture technology, and in particular to a method and system for digital culture product right confirmation and data sharing based on blockchain. Background Art

[0002] The rapid development of digital culture products in the Internet era faces challenges in right confirmation and data sharing. Traditional centralized right confirmation methods have problems such as low efficiency, easy data tampering, and lack of trust. At the same time, data sharing of digital culture products also faces difficulties in data security and privacy protection. How to ensure that data is not misused and leaked during the sharing process, and how to fairly distribute benefits are all problems that need to be solved urgently.

[0003] Traditional right confirmation methods rely on centralized institutions, with complex processes, cumbersome procedures, and high time costs.

[0004] The security of data sharing is difficult to guarantee, and there are risks of data leakage and abuse. Centralized data storage methods are easily targeted by hackers, resulting in data leakage. At the same time, there is a lack of effective access control mechanisms, making it difficult to ensure the security of data sharing.

[0005] There is a lack of a fair and reasonable income distribution mechanism, which is difficult to stimulate the creation and dissemination of digital culture products. Traditional income distribution models are usually opaque, and the rights and interests of creators are difficult to be effectively guaranteed, which is not conducive to the healthy development of the digital culture industry. Summary of the Invention

[0006] Embodiments of the present invention provide a method and system for digital culture product right confirmation and data sharing based on blockchain, which can solve the problems in the prior art.

[0007] In the first aspect of the embodiments of the present invention, A method for digital culture product right confirmation and data sharing based on blockchain is provided, including: Obtaining the original material information, creation process information, and ownership statement information submitted by the digital culture product creator; performing feature analysis on the original material information to extract multi-modal feature vectors; constructing a hierarchical neural network based on the multi-modal feature vectors, and using the hierarchical neural network to allocate the original material information to the basic right confirmation layer and generate a first feature matrix, allocate the creation process information to the derivative right confirmation layer and generate a second feature matrix, and allocate the ownership statement information to the collaborative right confirmation layer and generate a third feature matrix; Fusing the first feature matrix, the second feature matrix, and the third feature matrix to generate a comprehensive feature matrix of the digital culture product; constructing a self-evolving data trust network based on the comprehensive feature matrix, and establishing an encrypted data synchronization channel based on the self-evolving data trust network; Perform smart contract verification on the encrypted data synchronization channel; after successful verification, determine the access permission level of the encrypted data synchronization channel based on the comprehensive feature matrix; retrieve data at the corresponding level from the self-evolving data trust network according to the access permission level, and determine the temporal features of the data at the corresponding level; construct proof chain nodes based on the temporal features; use a graph neural network to establish a mapping relationship between the proof chain nodes and the encrypted data synchronization channel; perform homomorphic encryption on the mapping relationship to obtain encrypted features; use a federated learning algorithm to perform hierarchical aggregation calculation on the encrypted features to calculate the access contribution degree, and design a segmented revenue distribution curve according to the access contribution degree to execute revenue distribution.

[0008] Using the hierarchical neural network to allocate the original material information to the basic rights confirmation layer and generate a first feature matrix, allocate the creation process information to the derivative rights confirmation layer and generate a second feature matrix, and allocate the rights statement information to the collaborative rights confirmation layer and generate a third feature matrix includes: Input the original material information into the basic rights confirmation layer of the hierarchical neural network, and obtain a first feature matrix from the original material information through the basic rights confirmation layer using a feature extraction function, where the feature extraction function includes weight parameters, bias terms, and activation functions; Input the creation process information and the first feature matrix into the derivative rights confirmation layer of the hierarchical neural network, and perform feature fusion on the creation process information and the first feature matrix through the derivative rights confirmation layer to obtain a second feature matrix, where a first influence factor is introduced during the generation process of the second feature matrix to adjust the contribution weight of the first feature matrix; Input the rights statement information, the first feature matrix, and the second feature matrix into the collaborative rights confirmation layer of the hierarchical neural network, and perform feature fusion on the rights statement information, the first feature matrix, and the second feature matrix through the collaborative rights confirmation layer to obtain a third feature matrix, where a second influence factor and a third influence factor are introduced during the generation process of the third feature matrix to adjust the contribution weights of the first feature matrix and the second feature matrix respectively.

[0009] Fuse the first feature matrix, the second feature matrix, and the third feature matrix to generate a comprehensive feature matrix of the digital cultural product; construct a self-evolving data trust network based on the comprehensive feature matrix, and establish an encrypted data synchronization channel based on the self-evolving data trust network includes: Input the first feature matrix, the second feature matrix, and the third feature matrix into a reinforcement learning feature extraction network, which includes three parallel feature extraction layers; the reinforcement learning feature extraction network extracts features from the first feature matrix, the second feature matrix, and the third feature matrix respectively to obtain three feature vectors; input the three feature vectors into a feature fusion layer, and the feature fusion layer calculates the fusion weights of the three feature vectors through an attention mechanism; Output a Q-value estimate based on the fusion weights in a dual architecture of a target network and an evaluation network, and generate an optimal fusion weight vector according to the Q-value estimate; multiply and sum the first feature matrix, the second feature matrix, and the third feature matrix with the corresponding optimal fusion weights respectively to generate a comprehensive feature matrix of digital cultural products; Construct a self-evolving data trust network based on the comprehensive feature matrix, and establish an encrypted data synchronization channel based on the encryption public key in the self-evolving data trust network.

[0010] Verify the encrypted data synchronization channel through a smart contract; after successful verification, determine the access permission level of the encrypted data synchronization channel based on the comprehensive feature matrix; retrieve data at the corresponding level from the self-evolving data trust network according to the access permission level, and determine the timing characteristics of the data at the corresponding level, including: Receive an access application from a data requester, where the access application carries encrypted identity authentication information, and the encrypted identity authentication information includes the identity identification code of the data requester, an access timestamp, and a target data index number; Deploy an identity authentication smart contract, and the identity authentication smart contract calls the decryption key in the digital cultural product rights confirmation certificate to decrypt the encrypted identity authentication information; the identity authentication smart contract matches the decrypted identity identification code with the authorized user list in the digital cultural product rights confirmation certificate; the identity authentication smart contract determines whether the access timestamp is within the authorized access time period specified in the digital cultural product rights confirmation certificate; The identity authentication smart contract verifies whether the target data index number belongs to the authorized data range of the digital cultural product rights confirmation certificate; generate an identity authentication result based on the matching result of the identity identification code, the judgment result of the access timestamp, and the verification result of the target data index number; In the case where the identity authentication result is authentication passed, obtain the comprehensive trust score of the data requester, where the comprehensive trust score is calculated based on the historical access records, data usage compliance, and credit rating of the data requester; determine the access permission level of the data requester according to the comprehensive trust score, where the access permission level includes five levels, and each level corresponds to a different trust score interval; retrieve the data of the corresponding level from the self-evolving data trust network according to the access permission level, and determine the time series characteristics of the data of the corresponding level.

[0011] Constructing a proof chain node based on the time series characteristics includes: Extract the access time feature, access operation feature, and access object feature from the data of the corresponding level; convert the access time feature into a timestamp encoding, convert the access operation feature into a one-hot encoding, convert the access object feature into a vector encoding, and combine the timestamp encoding, the one-hot encoding, and the vector encoding to form an input feature vector; arrange the input feature vector in time series to construct an input feature sequence. Input the input feature sequence into a bidirectional long short-term memory network, where the bidirectional long short-term memory network includes a forward propagation network and a backward propagation network; the forward propagation network controls the input proportion of the input feature sequence at the current moment through an input gate, controls the forgetting proportion of historical features through a forgetting gate, and controls the output proportion of features through an output gate to generate a forward state vector; the backward propagation network adopts the same gating mechanism as the forward propagation network and processes the input feature sequence in the opposite time series to generate a backward state vector; splice the forward state vector and the backward state vector in the feature dimension to obtain the time series characteristics. Input the time series characteristics into a query feature transformation layer, a key feature transformation layer, and a value feature transformation layer respectively to generate a query feature matrix, a key feature matrix, and a value feature matrix; divide the query feature matrix, the key feature matrix, and the value feature matrix into multiple groups of attention features according to the number of attention heads; for each group of attention features, calculate the product of the query feature matrix and the transpose of the key feature matrix, divide the product by the square root value of the key feature matrix to obtain an attention score; normalize the attention score to obtain an attention weight coefficient; multiply the attention weight coefficient by the corresponding value feature matrix to obtain a weighted feature output; splice the weighted feature outputs in the feature dimension, and obtain a fusion feature vector through a feature fusion layer, and construct a proof chain node based on the fusion feature vector.

[0012] Establish a mapping relationship between the proof chain nodes and the encrypted data synchronization channel using a graph neural network; perform homomorphic encryption on the mapping relationship to obtain encrypted features; use a federated learning algorithm to perform hierarchical aggregation calculation on the encrypted features to obtain the access contribution degree, and design a segmented revenue distribution curve according to the access contribution degree to execute revenue distribution, including: Construct a graph neural network, and use the proof chain nodes and the digital cultural product rights confirmation vouchers as the input nodes of the graph neural network; use a feature transformation matrix to transform the feature vectors of the proof chain nodes and the digital cultural product rights confirmation vouchers respectively to obtain the initial node features; calculate the attention weights between nodes based on the initial node features in the graph neural network; Perform message passing and multi-layer aggregation on the node features according to the attention weights to obtain the updated node features; input the updated node features into a feature mapping layer to calculate the feature mapping matrix between the proof chain nodes and the digital cultural product rights confirmation vouchers; In the federated learning framework, use the feature mapping matrix to perform local feature extraction to obtain local encrypted features; perform hierarchical aggregation operations on the local encrypted features in the ciphertext domain to obtain global aggregation features; calculate the access quality score, timeliness score, and contribution value score based on the global aggregation features; perform weighted combination of the access quality score, the timeliness score, and the contribution value score with corresponding weights to obtain the final access contribution degree; Construct a three-segment revenue distribution curve, divide the revenue interval into a basic revenue interval, an incentive revenue interval, and a stable revenue interval according to the numerical range of the final access contribution degree; calculate the final revenue value in the corresponding revenue interval according to the final access contribution degree and execute revenue distribution.

[0013] Perform hierarchical aggregation operations on the local encrypted features in the ciphertext domain to obtain global aggregation features; calculate the access quality score, timeliness score, and contribution value score based on the global aggregation features, including: Perform vector addition operations on the local encrypted features in the ciphertext domain to obtain node aggregation features; calculate the node weight coefficients based on the node depth information and node historical contribution degree information of the node aggregation features, and perform weighted fusion of the node weight coefficients and the node aggregation features in the ciphertext domain to obtain inter-layer fusion features; Input the inter-layer fusion features into a theorem decomposition module to decompose the large number decryption task into multiple small number decryption subtasks; use the private key to perform parallel decryption on the small number decryption subtasks to obtain decryption sub-results; combine and reconstruct the decryption sub-results to obtain global aggregation features; Calculate the conditional entropy of the access sequence in the global aggregation feature to obtain the behavior entropy feature, extract the access depth feature and the access breadth feature in the global aggregation feature, and input the behavior entropy feature, the access depth feature, and the access breadth feature into a multi-layer perceptron model to obtain the access quality score; Calculate the adjacent access time interval based on the global aggregation feature, and input the adjacent access time interval into an exponential decay function to obtain a time decay factor; calculate the access frequency feature in the global aggregation feature; perform a weighted combination of the time decay factor and the access frequency feature to obtain the timeliness score; Construct a node influence graph, and calculate the node centrality index using the global aggregation feature; extract the out-degree information and in-degree information of the nodes in the node influence graph to obtain the node degree feature; extract the node historical reference information based on the global aggregation feature; perform weighting on the node centrality index, the node degree feature, and the node historical reference information to obtain the contribution value score.

[0014] In the second aspect of the embodiments of the present invention, Provide a blockchain-based digital cultural product right confirmation and data sharing system, including: A first unit for obtaining the original material information, the creation process information, and the ownership statement information submitted by the digital cultural product creator; performing feature analysis on the original material information to extract a multi-modal feature vector; constructing a hierarchical neural network based on the multi-modal feature vector, and using the hierarchical neural network to allocate the original material information to the basic right confirmation layer and generate a first feature matrix, allocate the creation process information to the derivative right confirmation layer and generate a second feature matrix, and allocate the ownership statement information to the collaborative right confirmation layer and generate a third feature matrix; A second unit for fusing the first feature matrix, the second feature matrix, and the third feature matrix to generate a comprehensive feature matrix of the digital cultural product; constructing a self-evolving data trust network based on the comprehensive feature matrix, and establishing an encrypted data synchronization channel based on the self-evolving data trust network; A third unit for performing smart contract verification on the encrypted data synchronization channel; after the verification is passed, determining the access permission level of the encrypted data synchronization channel based on the comprehensive feature matrix; retrieving the corresponding level of data from the self-evolving data trust network according to the access permission level, and determining the temporal characteristics of the corresponding level of data; constructing a proof chain node based on the temporal characteristics; establishing a mapping relationship between the proof chain node and the encrypted data synchronization channel using a graph neural network; performing homomorphic encryption on the mapping relationship to obtain an encrypted feature; using a federated learning algorithm to perform hierarchical aggregation calculation on the encrypted feature to calculate the access contribution degree, and designing a segmented revenue distribution curve according to the access contribution degree to perform revenue distribution.

[0015] The beneficial effects of this application are as follows: 1. It realizes the automation and intelligence of the confirmation of rights and data sharing for digital cultural products. The present invention uses a deep learning model to extract multi-modal features of digital cultural products and combines federated learning technology to generate a fingerprint of the confirmed rights features, thus realizing automated confirmation of rights. At the same time, based on an adaptive trust assessment model and intelligent contract verification, it realizes automated authorization and access control for data sharing.

[0016] 2. It improves the credibility and security of the confirmation of rights and data sharing for digital cultural products. The present invention adopts a hierarchical method for confirming rights, which conducts confirmation of rights from three levels: original materials, creation process, and ownership declaration, ensuring the integrity and reliability of the confirmation of rights. At the same time, the blockchain technology is used to store the confirmation vouchers, ensuring the immutability and transparency of the confirmation information. In addition, the encrypted data synchronization channel and homomorphic encryption technology in the data sharing process further enhance data security.

[0017] 3. It realizes the refined management of data sharing and fair income distribution. The present invention determines the access permission level of the data requester according to the comprehensive trust score, realizing the refined management of data sharing. At the same time, based on the access contribution degree, a segmented income distribution curve is designed to achieve fair income distribution, motivating data sharing and cultural creation. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a schematic flowchart of the method for confirming rights and sharing data of digital cultural products based on blockchain according to an embodiment of the present invention; Figure 2 is a schematic diagram comparing the accuracy of different methods for confirming rights on various digital cultural products according to an embodiment of the present invention; Figure 3 is a flowchart of intelligent contract verification for the encrypted data synchronization channel according to an embodiment of the present invention; Figure 4 is a schematic diagram of the income distribution effect under access simulation according to an embodiment of the present invention; Figure 5 is a flowchart of hierarchical aggregation and score calculation according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0020] The technical solution of the present invention will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0021] Figure 1 FIG. is a schematic flowchart of a method for digital cultural product rights confirmation and data sharing based on blockchain according to an embodiment of the present invention. As Figure 1 shown, the method includes: Obtain the original material information, creation process information, and ownership statement information submitted by the digital cultural product creator; perform feature analysis on the original material information to extract multi-modal feature vectors; construct a hierarchical neural network based on the multi-modal feature vectors, and use the hierarchical neural network to allocate the original material information to the basic rights confirmation layer and generate a first feature matrix, allocate the creation process information to the derivative rights confirmation layer and generate a second feature matrix, and allocate the ownership statement information to the collaborative rights confirmation layer and generate a third feature matrix; Fuse the first feature matrix, the second feature matrix, and the third feature matrix to generate a comprehensive feature matrix of the digital cultural product; construct a self-evolving data trust network based on the comprehensive feature matrix, and establish an encrypted data synchronization channel based on the self-evolving data trust network; Perform smart contract verification on the encrypted data synchronization channel; after the verification is passed, determine the access permission level of the encrypted data synchronization channel based on the comprehensive feature matrix; retrieve the corresponding level of data from the self-evolving data trust network according to the access permission level, and determine the timing characteristics of the corresponding level of data; construct a proof chain node based on the timing characteristics; establish a mapping relationship between the proof chain node and the encrypted data synchronization channel using a graph neural network; perform homomorphic encryption on the mapping relationship to obtain an encrypted feature; use a federated learning algorithm to perform hierarchical aggregation calculation on the encrypted feature to calculate the access contribution degree, and design a segmented revenue distribution curve according to the access contribution degree to execute revenue distribution.

[0022] In an alternative embodiment, using the hierarchical neural network to allocate the original material information to the basic rights confirmation layer and generate a first feature matrix, allocate the creation process information to the derivative rights confirmation layer and generate a second feature matrix, and allocate the ownership statement information to the collaborative rights confirmation layer and generate a third feature matrix includes: Input the original material information into the basic rights confirmation layer of the hierarchical neural network, and obtain a first feature matrix from the original material information through a feature extraction function by the basic rights confirmation layer. The feature extraction function includes weight parameters, bias terms, and activation functions; Input the creation process information and the first feature matrix into the derivative rights confirmation layer of the hierarchical neural network. Through the derivative rights confirmation layer, perform feature fusion on the creation process information and the first feature matrix to obtain a second feature matrix, where a first influence factor is introduced during the generation process of the second feature matrix to adjust the contribution weight of the first feature matrix. Input the ownership statement information, the first feature matrix, and the second feature matrix into the collaborative rights confirmation layer of the hierarchical neural network. Through the collaborative rights confirmation layer, perform feature fusion on the ownership statement information, the first feature matrix, and the second feature matrix to obtain a third feature matrix, where a second influence factor and a third influence factor are introduced during the generation process of the third feature matrix to adjust the contribution weights of the first feature matrix and the second feature matrix respectively.

[0023] The hierarchical neural network consists of three main levels: the basic rights confirmation layer, the derivative rights confirmation layer, and the collaborative rights confirmation layer. Each level independently completes specific functions, and at the same time, realizes information fusion between layers through feature transfer to form a complete rights confirmation system.

[0024] The basic rights confirmation layer receives the original material information as input. The original material information includes the basic constituent elements of digital cultural products, such as image data, audio data, text data, etc. Taking digital artworks as an example, the original material information can be a set of RGB image data with a size of 512×512 pixels, and each pixel point contains the values of three channels, with the value range from 0 to 255.

[0025] The basic rights confirmation layer uses a convolutional neural network structure for feature extraction. This structure contains five convolutional units, and each convolutional unit consists of a convolutional layer, a batch normalization layer, and an activation layer. The kernel sizes of the convolutional layers are 5×5, 3×3, 3×3, 3×3, and 3×3 in sequence, and the corresponding numbers of feature maps are 32, 64, 128, 256, and 512 respectively. The stride of the convolutional operation is 1, and the padding method is "same" to keep the size of the feature map. The ReLU function is used as the activation function, truncating the part of the input value less than 0 to 0 and keeping the part greater than 0 unchanged. The moving average decay rate of the batch normalization layer is set to 0.999, and the epsilon value is set to 0.001.

[0026] The weight parameters in the feature extraction function are set by the Xavier initialization method, which makes the weights follow a uniform distribution with a mean of 0 and a variance of 2 / (input dimension + output dimension). The initial value of the bias term is set to 0.1. To prevent overfitting, dropout layers are added after the last two convolutional units respectively, with dropout rates of 0.3 and 0.5.

[0027] After being processed by the basic rights confirmation layer, the original material information is converted into a first feature matrix with a dimension of 32×32×512. This matrix captures the key features of the original materials of digital cultural products, including visual elements such as texture, shape, and color distribution, providing basic information for the subsequent rights confirmation layer.

[0028] The derivative rights confirmation layer receives the creation process information and the first feature matrix as inputs. The creation process information includes creation tool information, creation time sequence information, and creator operation behavior information, etc. For example, for a digital painting work, the creation process information can include the type of brush used, the painting order, and the time stamps at each stage. The creation process information is encoded into a vector with a length of 256 after preprocessing.

[0029] The derivative rights confirmation layer uses an attention mechanism and a bidirectional long short-term memory network (Bi-LSTM) structure to fuse the creation process information and the first feature matrix. The attention mechanism generates an attention weight map by calculating the correlation between each element in the first feature matrix and the creation process information. The attention mechanism consists of three fully connected layers, with the number of hidden neurons being 256, 128, and 64. The ReLU activation function is used in the first two layers, and the Sigmoid activation function is used in the last layer to normalize the output to between 0 and 1.

[0030] The Bi-LSTM network consists of two LSTM layers in opposite directions, with each direction containing 128 hidden units. The forget gate, input gate, and output gate of the LSTM unit all use the Sigmoid activation function, while the candidate memory unit uses the tanh activation function. In addition, a first influence factor α is introduced as a trainable parameter to adjust the contribution weight of the first feature matrix. The initial value of α is set to 0.7 and is continuously optimized during the training process through the backpropagation algorithm.

[0031] After being processed by the derivative rights confirmation layer, a second feature matrix with a dimension of 16×16×256 is generated. This matrix contains the fusion result of the information features of the creation process and the original material features, and can reflect the evolution trajectory and creation characteristics of digital cultural products.

[0032] The collaborative rights confirmation layer receives the ownership statement information, the first feature matrix, and the second feature matrix as inputs. The ownership statement information includes creator identity information, copyright statement information, and authorization relationship information, etc. For example, for a digital music work, the ownership statement information can include composer information, performer information, and various copyright statements. The ownership statement information is encoded into a vector with a length of 192 after being processed.

[0033] The collaborative confirmation layer uses a graph neural network (GNN) structure to fuse multi-source information. The GNN contains three graph convolution layers with feature dimensions of 128, 96, and 64, respectively. The graph convolution operation uses a normalized adjacency matrix to implement message passing. Each graph convolution layer is followed by batch normalization and a PReLU activation function. The collaborative confirmation layer introduces the second influencing factor β and the third influencing factor γ as trainable parameters to adjust the contribution weights of the first feature matrix and the second feature matrix, respectively. The initial value of β is set to 0.3, and the initial value of γ is set to 0.6. These two parameters are also optimized during the training process through the back-propagation algorithm.

[0034] To enhance the feature fusion effect, the collaborative confirmation layer also introduces a residual connection mechanism and a feature gating unit. The residual connection directly adds the input feature to the corresponding output feature, effectively alleviating the gradient vanishing problem of the deep network. The feature gating unit controls the information flow of different feature channels through the sigmoid function, enabling the network to adaptively select important features.

[0035] After being processed by the collaborative confirmation layer, the third feature matrix is generated with a dimension of 8 × 8 × 128. This matrix combines the characteristics of the three information sources to form a complete feature representation of digital cultural products, providing a solid foundation for subsequent confirmation services.

[0036] Figure 2 This is a schematic diagram comparing the accuracy of different rights confirmation methods for various digital cultural products according to the embodiments of the present invention: The figure shows the performance comparison of three different technical solutions when processing different types of digital content. The figure shows the performance indicators of "this technical solution", "traditional hash coding method" and "CNN coding method" in five application scenarios of digital painting, digital music, digital games, digital literature and mixed types through line graphs. This technical solution performs best in all scenarios, and the performance indicators are always maintained above 90%: 95.3% in digital painting scenario, 93.8% in digital music, 96.4% in digital games, 95.2% in digital literature, and 92.1% in mixed types. The performance of CNN coding method is second, with performances of 88.7%, 87.2%, 85.6%, 87.3% and 81.9% in each scenario respectively. The traditional hash coding method performs the worst, with performance indicators of 79.5%, 78.4%, 76.8%, 80.6% and 75.3% in the five scenarios respectively. From the overall trend, the performance of the three solutions has declined when processing mixed types of content, but the performance advantage of this technical solution is still obvious, indicating that the solution has better generalization ability and stability. This performance comparison fully demonstrates the advancement and practical value of this technical solution in the field of digital content processing.

[0037] In an alternative embodiment, the first feature matrix, the second feature matrix, and the third feature matrix are fused to generate a comprehensive feature matrix of the digital cultural product; constructing a self-evolving data trust network based on the comprehensive feature matrix and establishing an encrypted data synchronization channel based on the self-evolving data trust network includes: Input the first feature matrix, the second feature matrix, and the third feature matrix into a reinforcement learning feature extraction network, which includes three parallel feature extraction layers; the reinforcement learning feature extraction network respectively extracts features from the first feature matrix, the second feature matrix, and the third feature matrix to obtain three feature vectors; input the three feature vectors into a feature fusion layer, and the feature fusion layer calculates the fusion weights of the three feature vectors through an attention mechanism; Output a Q-value estimate based on the fusion weights with a dual architecture of a target network and an evaluation network, generate an optimal fusion weight vector according to the Q-value estimate; multiply and sum the first feature matrix, the second feature matrix, and the third feature matrix respectively with the corresponding optimal fusion weights to generate a comprehensive feature matrix of the digital cultural product; Construct a self-evolving data trust network based on the comprehensive feature matrix, and establish an encrypted data synchronization channel based on the encryption public key in the self-evolving data trust network.

[0038] The process of fusing the first feature matrix, the second feature matrix, and the third feature matrix depends on a reinforcement learning feature extraction network. This network contains three parallel feature extraction layers, corresponding to the processing of the three types of feature matrices respectively. Specifically, the first feature matrix reflects the content features of the digital cultural product, with a dimension of m×n1; the second feature matrix represents the user interaction behavior features, with a dimension of m×n2; the third feature matrix contains time series features, with a dimension of m×n3.

[0039] The three parallel feature extraction layers of the reinforcement learning feature extraction network adopt a deep convolutional network structure. The first feature extraction layer contains 5 convolutional layers, each layer using 32, 64, 128, 256, and 512 convolutional kernels of size 3×3 respectively, and each layer is followed by a ReLU activation function and a batch normalization operation. The second feature extraction layer and the third feature extraction layer both adopt a similar structure, but adjust the number and size of the convolutional kernels to adapt to the dimensional characteristics of their respective input features.

[0040] After feature extraction, three feature vectors are obtained, denoted as v1, v2, and v3 respectively, and their dimensions are all 512. These three feature vectors are input into the feature fusion layer. The feature fusion layer uses a multi-head attention mechanism to calculate the fusion weights of the three feature vectors. In practical applications, 8 attention heads are adopted, and the dimension of each attention head is 64.

[0041] In the attention mechanism, for each feature vector vi, the query vector, key vector, and value vector are calculated: the feature vector is mapped to the query space, key space, and value space through three linear transformation matrices respectively. The dot product of the query vector and the key vector calculates the similarity, and after softmax normalization, the attention weights are obtained. The weighted sum of these weights and the value vector forms the output of each attention head. Finally, the outputs of all attention heads are concatenated and passed through a linear transformation to obtain the final output of the attention mechanism, which is the initial fusion weight vector.

[0042] A dual network architecture is adopted to optimize the fusion weights. The target network and the evaluation network have the same structure and are both composed of two fully connected layers. The first fully connected layer has 256 neurons and uses the ReLU activation function; the second fully connected layer has 128 neurons and uses the tanh activation function to ensure that the output value is between -1 and 1. The parameters of the evaluation network are updated in real time, and the parameters of the target network are updated every 100 training steps.

[0043] The evaluation network receives the current state (three feature vectors) and the action (the initial fusion weight vector) and outputs the Q-value estimate. After the state transition, the reward value r is calculated, which is based on the discriminative ability and data trustworthiness of the fused feature matrix. In the specific implementation, the reward value consists of two parts: one is the classification accuracy using cross-validation, and the other is the data consistency ratio verified by the blockchain.

[0044] Through the deep Q-learning algorithm, the optimization goal is to minimize the temporal difference error. Each training batch randomly samples 128 samples from the experience replay buffer. The Adam optimizer is used, the learning rate is set to 0.001, and the decay factor is 0.99. After 50,000 training steps, the optimal fusion weight vector is obtained, and its dimension is 3.

[0045] Multiply the three weight values in the optimal fusion weight vector by the three feature matrices respectively, and the specific operation is matrix scalar multiplication. Then, these three weighted feature matrices are added element-wise to obtain the comprehensive feature matrix F, whose dimension is the same as that of the original feature matrix. Taking a certain digital art as an example, the first feature matrix describes its visual elements, the second feature matrix reflects the user interaction pattern, and the third feature matrix captures the time evolution law. The optimized weights may be [0.45, 0.35, 0.2], indicating that the content features are more important in this scenario.

[0046] An self-evolving data trust network is constructed based on the comprehensive feature matrix F, and this network adopts an improved graph neural network architecture. The nodes represent digital cultural products, and the edges represent the degree of association between products. The cosine similarity between nodes is calculated through the comprehensive feature matrix, and connections are established between nodes with a similarity greater than 0.7. Each node has two types of attributes: one is the feature vector extracted from F, and the other is the data trust value of this node.

[0047] The initial trust value T(j) of node j is calculated based on historical interaction data. The trust propagation mechanism enables nodes to update their own trust values based on the trust scores of neighboring nodes. The update rule takes into account the weighted combination of direct and indirect trust values, and the update period is once every 24 hours. 10% of the nodes in the network are randomly selected as verification nodes, responsible for verifying the trust declarations of other nodes.

[0048] The self-evolution mechanism is achieved by periodically re-evaluating node connection relationships. Every 72 hours, the system recalculates the similarity between nodes based on the latest comprehensive feature matrix and adjusts the network topology accordingly. This ensures that the network can adapt to changes in the characteristics of digital cultural products and user preferences.

[0049] An encrypted data synchronization channel is established based on the self-evolving data trust network. Each node generates a pair of asymmetric encryption keys (public key pk, private key sk). The public key is stored in the trust network, and the private key is securely stored locally. When synchronizing data, the source node encrypts the data using the public key of the target node, and the target node decrypts the data using its own private key. For a specific digital art transaction, the sending node retrieves the public key of the receiving node, encrypts the transaction data using this public key, and then transmits it through the encrypted channel. The receiving node verifies the identity of the sending node and decrypts the data using its own private key.

[0050] In an alternative implementation, the encrypted data synchronization channel is verified by a smart contract; after passing the verification, the access permission level of the encrypted data synchronization channel is determined based on the comprehensive feature matrix; the corresponding level of data is retrieved from the self-evolving data trust network according to the access permission level, and the timing characteristics of the corresponding level of data are determined, including: Receiving an access application from a data requestor, the access application carrying encrypted identity authentication information, the encrypted identity authentication information including the identity identification code of the data requestor, an access timestamp, and a target data index number; Deploying an identity authentication smart contract, the identity authentication smart contract calling the decryption key in the digital cultural product rights confirmation certificate to decrypt the encrypted identity authentication information; the identity authentication smart contract matching the decrypted identity identification code with the authorized user list in the digital cultural product rights confirmation certificate; the identity authentication smart contract determining whether the access timestamp is within the authorized access time period specified in the digital cultural product rights confirmation certificate; The identity authentication smart contract verifying whether the target data index number belongs to the authorized data range of the digital cultural product rights confirmation certificate; generating an identity authentication result based on the matching result of the identity identification code, the judgment result of the access timestamp, and the verification result of the target data index number; In the case where the identity authentication result is authentication passed, obtain the comprehensive trust score of the data requester, where the comprehensive trust score is calculated based on the historical access records, data usage compliance, and credit rating of the data requester; determine the access permission level of the data requester according to the comprehensive trust score, where the access permission level includes five levels, and each level corresponds to a different trust score interval; retrieve the data at the corresponding level from the self-evolving data trust network according to the access permission level, and determine the time series characteristics of the data at the corresponding level.

[0051] First, receive the access application from the data requester. The access application contains encrypted identity authentication information, such as the identity identification code ID12345 of the data requester, the access timestamp 20240727100000, and the target data index number DC001. The encryption method uses the AES-256 algorithm, and the key is held by the data requester.

[0052] Next, deploy the identity authentication smart contract. The smart contract is stored on the blockchain, which is publicly transparent and immutable. The smart contract first calls the decryption key in the digital cultural product rights confirmation certificate to decrypt the encrypted identity authentication information. Suppose the decrypted identity identification code is ID12345, the access timestamp is 20240727100000, and the target data index number is DC001. Then, the smart contract matches the decrypted identity identification code ID12345 with the authorized user list in the digital cultural product rights confirmation certificate. Suppose the authorized user list contains ID12345, then the match is successful. Next, the smart contract determines whether the access timestamp 20240727100000 is within the authorized access time period specified in the digital cultural product rights confirmation certificate. Suppose the authorized access time period is from 20240727000000 to 20240727235959, then the timestamp is within the valid period. Finally, the smart contract verifies whether the target data index number DC001 belongs to the authorized data range of the digital cultural product rights confirmation certificate. Suppose the authorized data range contains DC001, then the verification passes. Based on the verification results of the above three conditions, the smart contract generates an identity authentication result, which is "authentication passed" in this case.

[0053] If the identity authentication result is "authentication passed", then obtain the comprehensive trust score of the data requester. The comprehensive trust score is calculated based on the historical access records, data usage compliance, and credit rating of the data requester. For example, if there has been no data leakage or abuse by the data requester in the past year, the data usage complies with relevant regulations, and the credit rating is A level, then the comprehensive trust score is 95 points.

[0054] Determine the access permission level of the data requester based on the comprehensive trust score. The access permission level is divided into five levels, and each level corresponds to a different trust score range. For example, 90 - 100 points is level one, 80 - 89 points is level two, 70 - 79 points is level three, 60 - 69 points is level four, and below 60 points is level five. In this case, the comprehensive trust score of the data requester is 95 points, corresponding to level one access permission.

[0055] Finally, retrieve the corresponding level of data from the self - evolving data trust network according to the access permission level. Level one permission can access all data, level two permission can access some sensitive data, level three permission can access public data, and level four and level five permissions can only access limited public data. In this case, the data requester has level one access permission and can access all data corresponding to the target data index number DC001.

[0056] The solution of this application can: Enhance data security: By encrypting identity authentication information, intelligent contract verification, and multi - level access permission control, it effectively prevents unauthorized access and data leakage, ensuring data security. Promote data circulation: By establishing a trust mechanism and refined access control, it promotes the circulation of data within a controllable range and maximizes the value of data. Improve access efficiency: The automated identity authentication and permission allocation process reduces manual intervention and improves data access efficiency.

[0057] Figure 3 This is the intelligent contract verification flowchart for encrypting the data synchronization channel in the embodiment of the present invention: This flowchart shows a complete data access request processing flow. The entire process starts with receiving an access application from the data requester, which includes information such as the identity identification code, access time period, and target data index number. Subsequently, it enters the link of deploying the identity authentication intelligent contract, which is divided into two parallel verification branches: the left branch is responsible for decrypting the identity authentication information and matching the identity identification code with the authorized user list; the right branch verifies whether the access time is within the authorized time range and verifies the authorization status of the target data index number. The verification results of the two branches are aggregated to form the identity authentication result. If the verification fails, the system will reject the access request and return a verification failure message; if the verification passes, it enters the subsequent processing flow: first, obtain the comprehensive trust score of the data requester (based on historical access records, usage compliance, and credit ratings), then determine the access permission level (divided into five levels for different trust score ranges), and finally retrieve the corresponding level of data from the self - evolving data trust network and determine the timing characteristics of the data. This process reflects the rigor and hierarchy of the system in data access control, ensuring the security and reliability of data access through multiple verification and dynamic evaluation mechanisms.

[0058] In an alternative embodiment, constructing a proof chain node based on the timing characteristics includes: Extracting access time characteristics, access operation characteristics, and access object characteristics from the data at the corresponding level; converting the access time characteristics into timestamp encoding, converting the access operation characteristics into one-hot encoding, converting the access object characteristics into vector encoding, and combining the timestamp encoding, the one-hot encoding, and the vector encoding to form an input feature vector; arranging the input feature vector in time sequence to construct an input feature sequence; Inputting the input feature sequence into a bidirectional long short-term memory network, where the bidirectional long short-term memory network includes a forward propagation network and a backward propagation network; the forward propagation network controls the input proportion of the input feature sequence at the current moment through an input gate, controls the forgetting proportion of historical features through a forgetting gate, and controls the output proportion of features through an output gate to generate a forward state vector; the backward propagation network adopts the same gating mechanism as the forward propagation network and processes the input feature sequence in the reverse time sequence to generate a backward state vector; splicing the forward state vector and the backward state vector in the feature dimension to obtain timing characteristics; Inputting the timing characteristics into a query feature transformation layer, a key feature transformation layer, and a value feature transformation layer respectively to generate a query feature matrix, a key feature matrix, and a value feature matrix; dividing the query feature matrix, the key feature matrix, and the value feature matrix into multiple groups of attention features according to the number of attention heads; for each group of attention features, calculating the product of the query feature matrix and the transpose of the key feature matrix, dividing the product by the square root value of the key feature matrix to obtain an attention score; normalizing the attention score to obtain an attention weight coefficient; multiplying the attention weight coefficient by the corresponding value feature matrix to obtain a weighted feature output; splicing the weighted feature outputs in the feature dimension and obtaining a fused feature vector through a feature fusion layer, and constructing a proof chain node based on the fused feature vector.

[0059] First, extract access time characteristics, access operation characteristics, and access object characteristics from the access log data at the corresponding level. For example, an access log record contains a user ID, an access time, an access operation (such as read, write, delete), and an access object (such as a file, a database record).

[0060] Then, encode the extracted features. Convert the access time feature into a timestamp encoding. For example, convert "2024-07-27 10:00:00" to a Unix timestamp. Convert the access operation feature into a one-hot encoding. For example, if the possible access operations include read, write, and delete, the "read" operation can be encoded as [1, 0, 0], the "write" operation can be encoded as [0, 1, 0], and the "delete" operation can be encoded as [0, 0, 1]. Convert the access object feature into a vector encoding. For example, the file name or database record ID can be converted into a vector representation using word embedding technology.

[0061] Next, combine the timestamp encoding, one-hot encoding, and vector encoding to form an input feature vector. For example, combine the timestamp 1690425600, the access operation one-hot encoding [1, 0, 0], and the access object vector encoding [0.2, 0.5, 0.8] into an input feature vector [1690425600, 1, 0, 0, 0.2, 0.5, 0.8].

[0062] Arrange multiple input feature vectors in the order of access time to construct an input feature sequence. For example, three consecutive accesses of a user can form an input feature sequence.

[0063] Input the input feature sequence into a bidirectional long short-term memory network. The bidirectional long short-term memory network consists of a forward propagation network and a backward propagation network. The forward propagation network processes the input feature sequence in chronological order, controls the flow of information through the input gate, forget gate, and output gate, and generates a forward state vector. The backward propagation network processes the input feature sequence in the reverse chronological order, also using the gating mechanism, and generates a backward state vector. For example, for an input sequence containing three feature vectors, the forward propagation network processes these three vectors in sequence, while the backward propagation network processes them in the reverse order.

[0064] Concatenate the forward state vector and the backward state vector in the feature dimension to obtain the temporal feature. For example, concatenate a forward state vector with a dimension of 128 and a backward state vector with a dimension of 128 to form a temporal feature with a dimension of 256.

[0065] Input the temporal feature into the query feature transformation layer, key feature transformation layer, and value feature transformation layer respectively to generate a query feature matrix, a key feature matrix, and a value feature matrix. These transformation layers can be simple linear layers.

[0066] Divide the query feature matrix, key feature matrix, and value feature matrix into multiple groups of attention features according to the number of attention heads. For example, if the number of attention heads is 8, each matrix is divided into 8 groups.

[0067] For each group of attention features, calculate the product of the query feature matrix and the transpose of the key feature matrix. Divide the product by the square root of the dimension of the key feature matrix to obtain the attention scores. Normalize the attention scores to obtain the attention weight coefficients. Multiply the attention weight coefficients by the corresponding value feature matrix to obtain the weighted feature output. Concatenate all the weighted feature outputs along the feature dimension and obtain the fused feature vector through a feature fusion layer (such as another linear layer).

[0068] Finally, construct the proof chain nodes based on the fused feature vector. For example, the fused feature vector can be used as the feature representation of the proof chain nodes.

[0069] The solution of this application can:[[]]END]] Improve accuracy: By combining the bidirectional long short-term memory network and the multi-head attention mechanism, it can more effectively capture the temporal dependence and feature correlation in the access log data, thereby improving the accuracy of proof chain construction. Enhance robustness: The bidirectional long short-term memory network can effectively process long sequence data, and the multi-head attention mechanism can focus on feature information in different aspects, making the method more robust to noise and outliers in the access log data. Improve efficiency: This method can automatically learn the feature representation in the access log data, avoiding cumbersome manual feature engineering, thus improving the efficiency of proof chain construction.

[0070] In an alternative embodiment, use a graph neural network to establish a mapping relationship between the proof chain nodes and the encrypted data synchronization channel; perform homomorphic encryption on the mapping relationship to obtain encrypted features; use a federated learning algorithm to perform hierarchical aggregation on the encrypted features to calculate the access contribution degree, and design a segmented revenue distribution curve according to the access contribution degree to perform revenue distribution, including: Construct a graph neural network, and use the proof chain nodes and the digital cultural product rights confirmation certificate as the input nodes of the graph neural network; use the feature transformation matrix to transform the feature vectors of the proof chain nodes and the digital cultural product rights confirmation certificate respectively to obtain the initial node features; calculate the attention weights between nodes based on the initial node features in the graph neural network. Perform message passing and multi-layer aggregation on the node features according to the attention weights to obtain the updated node features; input the updated node features into the feature mapping layer to calculate the feature mapping matrix between the proof chain nodes and the digital cultural product rights confirmation certificate. In the federated learning framework, local encrypted features are obtained by using the feature mapping matrix for local feature extraction; hierarchical aggregation operations are performed on the local encrypted features in the ciphertext domain to obtain global aggregation features; access quality scores, timeliness scores, and contribution value scores are calculated based on the global aggregation features; and the access quality scores, the timeliness scores, and the contribution value scores are weighted and combined with corresponding weights to obtain the final access contribution degree. A three-segment revenue distribution curve is constructed, and the revenue interval is divided into a basic revenue interval, an incentive revenue interval, and a stable revenue interval according to the numerical range of the final access contribution degree; and the final revenue value is calculated within the corresponding revenue interval according to the final access contribution degree and revenue distribution is performed.

[0071] First, a graph neural network is constructed. Nodes on the proof chain, such as transaction records, creation records, etc., and digital cultural product rights confirmation vouchers, such as digital signatures, timestamps, etc., are used as input nodes of the graph neural network. Feature extraction is performed on these nodes. For example, for proof chain nodes, features such as time, transaction amount, participants, etc. can be extracted; for rights confirmation vouchers, features such as creation time, certificate number, certification agency, etc. can be extracted. These features are converted into initial feature vectors of the nodes by using a feature transformation matrix. In the graph neural network, the attention weights between nodes are calculated based on the initial feature vectors of the nodes. For example, the more similar the features of two nodes are, the higher their attention weights are. Then, message passing is performed according to the attention weights to transfer the feature information of the nodes to adjacent nodes. After multiple layers of aggregation, the updated node features are obtained. Finally, the updated node features are input into the feature mapping layer to calculate the feature mapping matrix between the proof chain nodes and the digital cultural product rights confirmation vouchers, and this matrix reflects the association strength between them.

[0072] For example, suppose there are two proof chain nodes A and B, and a rights confirmation voucher C. The feature vector of A is [0.2, 0.5, 0.3], the feature vector of B is [0.1, 0.6, 0.3], and the feature vector of C is [0.3, 0.4, 0.3]. After calculation, the attention weight between A and C is 0.8, and the attention weight between B and C is 0.7. After message passing and aggregation, the feature vector of A is updated to [0.25, 0.45, 0.3], and the feature vector of B is updated to [0.15, 0.55, 0.3]. The final calculated feature mapping value between A and C is 0.9, and the feature mapping value between B and C is 0.8.

[0073] Next, generate a homomorphic encryption key pair. Select two large prime numbers, calculate their product as part of the public key, and select a generator. The least common multiple of the two prime numbers minus one is used as the private key. Encrypt each element in the feature mapping matrix using the public key. The encryption process includes exponentiation of the generator and modular operation of a random number. Obtain the encrypted mapping matrix, where each element is a ciphertext.

[0074] For example, select prime numbers p = 7 and q = 11, then part of the public key n = 77, and select the generator g = 2. The private key is (7 - 1) * (11 - 1) / greatest common divisor(6, 10) = 30. Encrypt the feature mapping value 0.9. Assuming the random number is 3, the encryption result is (2^0.9 * 3) mod 77 = 55.

[0075] Under the federated learning framework, each participant uses the encrypted mapping matrix for local feature extraction to obtain local encrypted features. Then, perform hierarchical aggregation operations on the local encrypted features in the ciphertext domain to obtain global aggregated features. Decrypt the global aggregated features using the private key.

[0076] For example, two participants calculate local encrypted features 55 and 60 respectively. Perform an addition operation in the ciphertext domain to obtain the global aggregated feature 115 mod 77 = 38. Decrypt using the private key to obtain the final aggregated feature.

[0077] Based on the decrypted global aggregated features, calculate the access quality score, timeliness score, and contribution value score. For example, the access quality score can be calculated based on the user's activity and reputation; the timeliness score can be calculated based on the access time and frequency; the contribution value score can be calculated based on the valid information and feedback provided by the user. Combine these scores with preset weights to obtain the final access contribution degree.

[0078] Assume the access quality score is 0.8, the timeliness score is 0.9, and the contribution value score is 0.7, with corresponding weights of 0.4, 0.3, and 0.3 respectively. Then the final access contribution degree is 0.8 * 0.4 + 0.9 * 0.3 + 0.7 * 0.3 = 0.8.

[0079] Finally, the revenue distribution is carried out according to the final access contribution. A three - stage revenue distribution curve is constructed, dividing the revenue range into a basic revenue range, an incentive revenue range, and a stable revenue range. In the basic revenue range, a linear function is used to calculate the revenue, and the slope of the linear function increases with the increase in participation. In the incentive revenue range, a quadratic function is used to calculate the revenue, and the coefficients of the quadratic function are adjusted according to the innovation degree. In the stable revenue range, a logarithmic function is used to calculate the revenue, and the base of the logarithmic function is determined based on the system stability. At the demarcation points between adjacent revenue ranges, a smooth transition function, such as weighted average, is introduced to ensure the continuity of revenue distribution. The final revenue value is calculated within the corresponding revenue range according to the final access contribution and the revenue distribution is executed.

[0080] For example, assume that the final access contribution is 0.8, which falls within the incentive revenue range. The revenue is calculated according to the quadratic function, and assume the revenue is 10.

[0081] Figure 4 The following is a schematic diagram of the revenue distribution effect under the access simulation of the embodiments of the present invention: This figure shows the performance comparison of three different revenue distribution schemes under different access scenarios. In the figure, a box - and - whisker plot is used to show the revenue distribution of "the technical solution of the present application (three - stage revenue)", "linear revenue distribution", and "fixed revenue distribution" in three scenarios of high - quality access, standard access, and low - quality access. In the high - quality access scenario, the maximum revenue of the technical solution of the present application can reach 135 coins; the median revenue of the linear revenue distribution scheme is about 98 coins; the fixed revenue always remains at 50 coins. In the standard access scenario, the median revenue of the technical solution of the present application is about 85 coins, the median of the linear revenue distribution is 62 coins, and the fixed revenue still maintains at the level of 50 coins. In the low - quality access scenario, the revenues of the technical solution of the present application and the linear revenue distribution both drop to about 24 coins, lower than the 50 coins of the fixed revenue. This hierarchical revenue distribution mechanism fully reflects the incentive characteristics of the technical solution of the present application: higher rewards are given for high - quality access, and the revenue is punitively reduced for low - quality access, thus effectively guiding users to provide high - quality access behaviors and optimizing the overall service quality of the system.

[0082] The solution of this application can: The confirmation of rights and traceability are more reliable. By using graph neural networks and blockchain technology, the source and transfer process of digital cultural products can be effectively traced, the rights and interests of digital cultural products can be protected, and piracy and infringement can be prevented. The revenue distribution is more fair and reasonable. The revenue distribution is based on the contribution degree, which encourages users to actively participate in the creation and dissemination of digital cultural products and promotes the healthy development of the digital cultural industry. The system is more secure and stable. By adopting homomorphic encryption technology, the data privacy and security of users are protected. At the same time, the federated learning framework avoids the risks brought by centralized data storage and improves the stability and reliability of the system.

[0083] In an alternative embodiment, a hierarchical aggregation operation is performed on the local encryption features in the ciphertext domain to obtain global aggregation features; calculating the access quality score, timeliness score, and contribution value score based on the global aggregation features includes: Performing a vector addition operation on the local encryption features in the ciphertext domain to obtain node aggregation features; calculating node weight coefficients based on the node depth information and node historical contribution degree information of the node aggregation features, and performing weighted fusion on the node weight coefficients and the node aggregation features in the ciphertext domain to obtain inter-layer fusion features; Inputting the inter-layer fusion features into a theorem decomposition module to decompose the large number decryption task into multiple small number decryption subtasks; performing parallel decryption on the small number decryption subtasks using a private key to obtain decryption sub-results; combining and reconstructing the decryption sub-results to obtain global aggregation features; Calculating the conditional entropy of the access sequence in the global aggregation features to obtain a behavior entropy feature, extracting the access depth feature and access breadth feature in the global aggregation features, and inputting the behavior entropy feature, the access depth feature, and the access breadth feature into a multi-layer perceptron model to obtain an access quality score; Calculating the adjacent access time interval based on the global aggregation features, and inputting the adjacent access time interval into an exponential decay function to obtain a time decay factor; calculating the access frequency feature in the global aggregation features; performing weighted combination on the time decay factor and the access frequency feature to obtain a timeliness score; Constructing a node influence graph, and calculating node centrality metrics using the global aggregation features; extracting the out-degree information and in-degree information of the nodes in the node influence graph to obtain node degree features; extracting node historical reference information based on the global aggregation features; performing weighting on the node centrality metrics, the node degree features, and the node historical reference information to obtain a contribution value score.

[0084] First, collect the user's local access data, such as the accessed web page link, access time, access duration, etc. Extract features from the local access data of each user, such as access sequence, access depth, access breadth, etc., and convert these features into vector representations. Use a homomorphic encryption algorithm to encrypt the feature vectors of each user to obtain local encryption features.

[0085] Next, perform vector addition operations on the local encryption features of nodes at the same level to obtain node aggregation features. For example, add the local encryption features of all users in the same department. Calculate the node weight coefficient according to the depth information of the node (e.g., the level in the organizational structure) and the historical contribution degree information of the node (e.g., the quantity and quality of the content contributed in the past). Perform weighted fusion of the node weight coefficient and the node aggregation feature in the ciphertext domain to obtain the inter-layer fusion feature. For example, higher-level nodes and nodes with a high historical contribution degree are assigned higher weights. Suppose there are two nodes, the aggregation feature of node A is [1, 2, 3] and the weight is 0.8; the aggregation feature of node B is [4, 5, 6] and the weight is 0.2. Then the inter-layer fusion feature is [1*0.8 + 4*0.2, 2*0.8 + 5*0.2, 3*0.8 + 6*0.2] = [1.6, 2.6, 3.6].

[0086] Then, input the inter-layer fusion feature into the theorem decomposition module. This module decomposes the large number decryption task into multiple small number decryption subtasks. For example, decompose a 1024-bit ciphertext into 16 64-bit ciphertexts. Use the private key to decrypt these small number decryption subtasks in parallel to obtain decryption sub-results. Combine and reconstruct the decryption sub-results to obtain the global aggregation feature.

[0087] After obtaining the global aggregation feature, calculate the access quality score. Calculate the conditional entropy of the access sequence in the global aggregation feature to obtain the behavior entropy feature. For example, the more random the access sequence, the higher the behavior entropy. Extract the access depth feature and the access breadth feature from the global aggregation feature. For example, the deeper the level of the web page visited, the higher the access depth; the more types of web pages visited, the higher the access breadth. Input the behavior entropy feature, the access depth feature, and the access breadth feature into the multi-layer perceptron model to obtain the access quality score.

[0088] Next, calculate the timeliness score. Calculate the adjacent access time interval based on the global aggregation feature. Input the adjacent access time interval into the exponential decay function to obtain the time decay factor. For example, the longer the time interval, the smaller the time decay factor. Calculate the access frequency feature in the global aggregation feature. For example, the more access times, the higher the access frequency. Perform weighted combination of the time decay factor and the access frequency feature to obtain the timeliness score.

[0089] Finally, calculate the contribution value score. Construct a node influence graph and calculate the node centrality index using global aggregation features. For example, the more connections a node has, the higher its centrality. Extract the out-degree information (the number of outgoing connections) and in-degree information (the number of incoming connections) of the nodes in the node influence graph to obtain the node degree features. Extract the historical citation information of the nodes based on the global aggregation features. For example, the more times the content of a node is cited, the higher its historical citation information. Input the node centrality index, node degree features, and node historical citation information into the attention network to obtain the feature weights. Weight the node centrality index, node degree features, and node historical citation information based on the feature weights to obtain the contribution value score.

[0090] Figure 5 Flowchart of hierarchical aggregation and score calculation in an embodiment of the present invention: This flowchart shows a complex feature calculation and fusion process that starts from local encrypted features and finally obtains the contribution value score after multiple processing stages. The process first performs vector addition operations in the ciphertext domain to obtain node aggregation features, and at the same time calculates the node weight coefficients based on the node depth information and historical contribution degree information. These two parts of information are weighted and fused in the ciphertext domain to obtain the inter-layer fusion features. Subsequently, the system inputs the inter-layer fusion features into the theorem decomposition module, decomposes the large number decryption task into multiple small number decryption subtasks, and performs parallel decryption on these subtasks through the private key to obtain the decryption sub-results. The decryption sub-results are combined and reconstructed to obtain the global aggregation features, and in-depth analysis is carried out in three directions on this basis: one is to calculate the conditional entropy of the access sequence to obtain the behavior features; the second is to extract the access depth features and access breadth features; the third is to calculate the access time interval and introduce an exponential decay factor to obtain the timeliness score. These features are further processed by a multi-layer perceptron model, and finally a node influence graph is constructed, the out-degree information and in-degree information of the nodes are extracted, and combined with the node centrality index, node degree features, and node historical citation information, and the final contribution value score is obtained through weighted calculation. The entire process reflects the integrity and scientificity of the system in data processing, secure computing, and value evaluation.

[0091] The solution of this application can: Improve efficiency: Through homomorphic encryption and theorem decomposition technologies, efficient aggregation operations can be realized in the ciphertext domain, avoiding the leakage of plaintext data and improving the computing efficiency at the same time. Enhance accuracy: Through deep learning technologies such as multi-layer perceptron models and attention networks, the access quality, timeliness, and contribution value of users can be evaluated more accurately. Enhance security: Use homomorphic encryption technology to protect user data, effectively preventing the leakage of sensitive information and ensuring the security of user data.

[0092] In the second aspect of the embodiment of the present invention, Provided is a blockchain-based digital cultural product rights confirmation and data sharing system, including: The first unit is used to obtain the original material information, creation process information, and ownership statement information submitted by the digital cultural product creator; perform feature analysis on the original material information to extract multi-modal feature vectors; construct a hierarchical neural network based on the multi-modal feature vectors, and use the hierarchical neural network to allocate the original material information to the basic rights confirmation layer and generate a first feature matrix, allocate the creation process information to the derivative rights confirmation layer and generate a second feature matrix, and allocate the ownership statement information to the collaborative rights confirmation layer and generate a third feature matrix; The second unit is used to fuse the first feature matrix, the second feature matrix, and the third feature matrix to generate a comprehensive feature matrix of the digital cultural product; construct a self-evolving data trust network based on the comprehensive feature matrix, and establish an encrypted data synchronization channel based on the self-evolving data trust network; The third unit is used to perform smart contract verification on the encrypted data synchronization channel; after the verification passes, determine the access permission level of the encrypted data synchronization channel based on the comprehensive feature matrix; retrieve the corresponding level of data from the self-evolving data trust network according to the access permission level, and determine the temporal characteristics of the corresponding level of data; construct a proof chain node based on the temporal characteristics; use a graph neural network to establish a mapping relationship between the proof chain node and the encrypted data synchronization channel; perform homomorphic encryption on the mapping relationship to obtain an encrypted feature; use a federated learning algorithm to perform hierarchical aggregation calculation on the encrypted feature to calculate the access contribution degree, and design a segmented revenue distribution curve according to the access contribution degree to execute revenue distribution.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and 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 method for the confirmation of rights and data sharing of digital cultural products based on blockchain, characterized in that, Including: Obtain the original material information, creation process information, and ownership statement information submitted by the digital cultural product creator; Conduct feature analysis on the original material information to extract multi-modal feature vectors; construct a hierarchical neural network based on the multi-modal feature vectors, use the hierarchical neural network to allocate the original material information to the basic rights confirmation layer and generate a first feature matrix, allocate the creation process information to the derivative rights confirmation layer and generate a second feature matrix, and allocate the ownership statement information to the collaborative rights confirmation layer and generate a third feature matrix; Fuse the first feature matrix, the second feature matrix, and the third feature matrix to generate a comprehensive feature matrix of the digital cultural product; Construct a self-evolving data trust network based on the comprehensive feature matrix, and establish an encrypted data synchronization channel based on the self-evolving data trust network; Perform intelligent contract verification on the encrypted data synchronization channel; after passing the verification, determine the access permission level of the encrypted data synchronization channel based on the comprehensive feature matrix; retrieve the corresponding level of data from the self-evolving data trust network according to the access permission level, and determine the temporal characteristics of the corresponding level of data; Construct a proof chain node based on the temporal characteristics; use a graph neural network to establish a mapping relationship between the proof chain node and the encrypted data synchronization channel; Perform homomorphic encryption on the mapping relationship to obtain an encrypted feature; use a federated learning algorithm to perform hierarchical aggregation calculation on the encrypted feature to obtain the access contribution degree, and design a segmented revenue distribution curve according to the access contribution degree to execute revenue distribution.

2. The method according to claim 1, wherein Using the hierarchical neural network to allocate the original material information to the basic rights confirmation layer and generate a first feature matrix, allocate the creation process information to the derivative rights confirmation layer and generate a second feature matrix, and allocate the ownership statement information to the collaborative rights confirmation layer and generate a third feature matrix includes: Input the original material information into the basic rights confirmation layer of the hierarchical neural network, and obtain a first feature matrix from the original material information through a feature extraction function in the basic rights confirmation layer, where the feature extraction function includes weight parameters, bias terms, and activation functions; Input the creation process information and the first feature matrix into the derivative rights confirmation layer of the hierarchical neural network, and perform feature fusion on the creation process information and the first feature matrix through the derivative rights confirmation layer to obtain a second feature matrix, where a first influence factor is introduced in the generation process of the second feature matrix to adjust the contribution weight of the first feature matrix; Input the ownership statement information, the first feature matrix, and the second feature matrix into the collaborative rights confirmation layer of the hierarchical neural network, and perform feature fusion on the ownership statement information, the first feature matrix, and the second feature matrix through the collaborative rights confirmation layer to obtain a third feature matrix, where a second influence factor and a third influence factor are introduced in the generation process of the third feature matrix to adjust the contribution weights of the first feature matrix and the second feature matrix respectively.

3. The method according to claim 1, wherein Fuse the first feature matrix, the second feature matrix, and the third feature matrix to generate a comprehensive feature matrix of the digital cultural product; Constructing a self-evolving data trust network based on the comprehensive feature matrix and establishing an encrypted data synchronization channel based on the self-evolving data trust network includes: Inputting the first feature matrix, the second feature matrix, and the third feature matrix into a reinforcement learning feature extraction network, where the reinforcement learning feature extraction network includes three parallel feature extraction layers; the reinforcement learning feature extraction network respectively extracts features from the first feature matrix, the second feature matrix, and the third feature matrix to obtain three feature vectors; inputting the three feature vectors into a feature fusion layer, and the feature fusion layer calculates the fusion weights of the three feature vectors through an attention mechanism; Outputting a Q-value estimate based on the fusion weights with a dual architecture of a target network and an evaluation network, generating an optimal fusion weight vector according to the Q-value estimate; multiplying and summing the first feature matrix, the second feature matrix, and the third feature matrix respectively with the corresponding optimal fusion weights to generate a comprehensive feature matrix of digital cultural products; Constructing a self-evolving data trust network based on the comprehensive feature matrix and establishing an encrypted data synchronization channel based on the encrypted public key in the self-evolving data trust network.

4. The method according to claim 1, wherein Performing intelligent contract verification on the encrypted data synchronization channel; after passing the verification, determining the access permission level of the encrypted data synchronization channel based on the comprehensive feature matrix; Retrieving data at the corresponding level from the self-evolving data trust network according to the access permission level, and determining the temporal characteristics of the data at the corresponding level, including: Receiving an access application from a data requester, where the access application carries encrypted identity authentication information, and the encrypted identity authentication information includes the identity identification code of the data requester, an access timestamp, and a target data index number; Deploying an identity authentication intelligent contract, where the identity authentication intelligent contract calls the decryption key in the digital cultural product rights confirmation certificate to decrypt the encrypted identity authentication information; the identity authentication intelligent contract matches the decrypted identity identification code with the authorized user list in the digital cultural product rights confirmation certificate; the identity authentication intelligent contract determines whether the access timestamp is within the authorized access time period specified in the digital cultural product rights confirmation certificate; The identity authentication intelligent contract verifies whether the target data index number belongs to the authorized data range of the digital cultural product rights confirmation certificate; generating an identity authentication result based on the matching result of the identity identification code, the judgment result of the access timestamp, and the verification result of the target data index number; In the case where the identity authentication result is authentication passed, obtaining the comprehensive trust score of the data requester, where the comprehensive trust score is calculated based on the historical access records, data usage compliance, and credit rating of the data requester; determining the access permission level of the data requester according to the comprehensive trust score, and the access permission level includes five levels, and each level corresponds to a different trust score interval; retrieving data at the corresponding level from the self-evolving data trust network according to the access permission level, and determining the temporal characteristics of the data at the corresponding level.

5. The method according to claim 1, characterized in that, Constructing the proof chain nodes based on the timing characteristics includes: Extracting the access time feature, access operation feature, and access object feature from the data at the corresponding level; converting the access time feature into a timestamp encoding, converting the access operation feature into a one-hot encoding, converting the access object feature into a vector encoding, and combining the timestamp encoding, the one-hot encoding, and the vector encoding to form an input feature vector; arranging the input feature vector in time sequence to construct an input feature sequence; Inputting the input feature sequence into a bidirectional long short-term memory network, where the bidirectional long short-term memory network includes a forward propagation network and a backward propagation network; the forward propagation network controls the input proportion of the input feature sequence at the current moment through an input gate, controls the forgetting proportion of historical features through a forgetting gate, and controls the output proportion of features through an output gate to generate a forward state vector; the backward propagation network uses the same gating mechanism as the forward propagation network to process the input feature sequence in the reverse time sequence to generate a backward state vector; splicing the forward state vector and the backward state vector in the feature dimension to obtain a timing feature; Inputting the timing feature into a query feature transformation layer, a key feature transformation layer, and a value feature transformation layer respectively to generate a query feature matrix, a key feature matrix, and a value feature matrix; equally dividing the query feature matrix, the key feature matrix, and the value feature matrix into multiple groups of attention features according to the number of attention heads; for each group of the attention features, calculating the product of the query feature matrix and the transpose of the key feature matrix, dividing the product by the square root value of the key feature matrix to obtain an attention score; normalizing the attention score to obtain an attention weight coefficient; multiplying the attention weight coefficient by the corresponding value feature matrix to obtain a weighted feature output; splicing the weighted feature outputs in the feature dimension, and obtaining a fused feature vector through a feature fusion layer, and constructing a proof chain node based on the fused feature vector.

6. The method according to claim 1, wherein Using a graph neural network to establish a mapping relationship between the proof chain node and the encrypted data synchronization channel; performing homomorphic encryption on the mapping relationship to obtain an encrypted feature; Adopting a federated learning algorithm to perform hierarchical aggregation calculation on the encrypted feature to obtain the access contribution degree, and designing a segmented revenue distribution curve according to the access contribution degree to execute revenue distribution, including: Constructing a graph neural network, and taking the proof chain node and the digital cultural product rights confirmation certificate as input nodes of the graph neural network; respectively converting the feature vectors of the proof chain node and the digital cultural product rights confirmation certificate by using a feature transformation matrix to obtain node initial features; calculating the attention weights between nodes based on the node initial features in the graph neural network; Performing message passing and multi-layer aggregation on the node features according to the attention weights to obtain updated node features; inputting the updated node features into a feature mapping layer to calculate the feature mapping matrix between the proof chain node and the digital cultural product rights confirmation certificate; In the federated learning framework, local encrypted features are obtained by using the feature mapping matrix for local feature extraction; hierarchical aggregation operations are performed on the local encrypted features in the ciphertext domain to obtain global aggregated features; access quality scores, timeliness scores, and contribution value scores are calculated based on the global aggregated features; the access quality scores, the timeliness scores, and the contribution value scores are weighted and combined with corresponding weights to obtain the final access contribution degree. A three-stage revenue distribution curve is constructed, and the revenue interval is divided into a basic revenue interval, an incentive revenue interval, and a stable revenue interval according to the numerical range of the final access contribution degree; the final revenue value is calculated within the corresponding revenue interval according to the final access contribution degree and revenue distribution is performed.

7. The method according to claim 6, wherein Hierarchical aggregation operations are performed on the local encrypted features in the ciphertext domain to obtain global aggregated features. Calculating the access quality score, timeliness score, and contribution value score based on the global aggregated features includes: Performing vector addition operations on the local encrypted features in the ciphertext domain to obtain node aggregated features; calculating node weight coefficients based on the node depth information and node historical contribution degree information of the node aggregated features, and performing weighted fusion of the node weight coefficients and the node aggregated features in the ciphertext domain to obtain inter-layer fusion features. Inputting the inter-layer fusion features into a theorem decomposition module to decompose the large number decryption task into multiple small number decryption subtasks; performing parallel decryption on the small number decryption subtasks using a private key to obtain decryption sub-results; combining and reconstructing the decryption sub-results to obtain global aggregated features. Calculating the conditional entropy of the access sequence in the global aggregated features to obtain a behavior entropy feature, extracting the access depth feature and access breadth feature in the global aggregated features, and inputting the behavior entropy feature, the access depth feature, and the access breadth feature into a multi-layer perceptron model to obtain the access quality score. Calculating the adjacent access time interval based on the global aggregated features, inputting the adjacent access time interval into an exponential decay function to obtain a time decay factor; calculating the access frequency feature in the global aggregated features; performing weighted combination of the time decay factor and the access frequency feature to obtain the timeliness score. Constructing a node influence graph, calculating node centrality indicators using the global aggregated features; extracting the out-degree information and in-degree information of the nodes in the node influence graph to obtain node degree features; extracting node historical citation information based on the global aggregated features; weighting the node centrality indicators, the node degree features, and the node historical citation information to obtain the contribution value score.

8. A blockchain-based digital cultural product right confirmation and data sharing system for implementing the method described in any one of the foregoing claims 1-7, characterized in that Including: A first unit for obtaining the original material information, creation process information, and ownership statement information submitted by the digital cultural product creator. Performing feature analysis on the original material information to extract multi-modal feature vectors; constructing a hierarchical neural network based on the multi-modal feature vectors, using the hierarchical neural network to allocate the original material information to the basic rights confirmation layer and generate a first feature matrix, allocate the creation process information to the derivative rights confirmation layer and generate a second feature matrix, and allocate the ownership statement information to the collaborative rights confirmation layer and generate a third feature matrix. A second unit, configured to fuse the first feature matrix, the second feature matrix, and the third feature matrix to generate a comprehensive feature matrix of the digital cultural product; Construct a self-evolving data trust network according to the comprehensive feature matrix, and establish an encrypted data synchronization channel based on the self-evolving data trust network; A third unit, configured to perform smart contract verification on the encrypted data synchronization channel; after the verification passes, determine the access privilege level of the encrypted data synchronization channel based on the comprehensive feature matrix; retrieve data of the corresponding level from the self-evolving data trust network according to the access privilege level, and determine the timing characteristics of the data of the corresponding level; Construct a proof chain node based on the timing characteristics; use a graph neural network to establish a mapping relationship between the proof chain node and the encrypted data synchronization channel; Perform homomorphic encryption on the mapping relationship to obtain an encrypted feature; Adopt a federated learning algorithm to perform hierarchical aggregation calculation on the access contribution degree of the encrypted feature, and design a segmented revenue distribution curve according to the access contribution degree to execute revenue distribution.

Images