Blockchain encryption and deep learning retrieval methods for metaverse e-commerce platforms

Abstract

The present invention discloses a method for blockchain encryption and deep learning retrieval for a metaverse e-commerce platform, which protects the privacy of user behavior data, virtual product features, and social network relationships through homomorphic encryption technology and zero-knowledge proof protocol, generates tamper-proof transaction vouchers, and constructs a multimodal encrypted data set; designs a deep learning retrieval model with an attention mechanism and optimizes its parameters; generates a lightweight model through quantitative pruning processing, deploys it on distributed edge nodes, and realizes real-time rendering and low-latency interaction of virtual products in combination with a dynamic resource scheduling algorithm; introduces a federated learning framework to incrementally update the model; establishes a cloud-edge collaborative blockchain synchronization mechanism, and uses sharding technology and consensus algorithms to ensure data consistency. The present invention solves the problems of privacy leakage, insufficient 3D feature analysis, high latency in high-concurrency interactions, and low data collaboration efficiency in traditional metaverse e-commerce platforms, and significantly improves the overall performance of the platform.

Description

Technical Field

[0001] The present invention belongs to the field of the intersection of e-commerce and information technology, and specifically provides a method for blockchain encryption and deep learning retrieval for the Metaverse e-commerce platform. Background Art

[0002] With the rapid development of metaverse technology, metaverse e-commerce platforms have gradually become a key vehicle for integrating the virtual and real economies. Users engage in immersive interactions for product browsing, real-time transactions, and social activities, posing multi-faceted challenges to the platform's technical architecture. However, existing technical solutions suffer from significant shortcomings in core areas such as privacy protection, feature analysis, resource scheduling, and data collaboration, severely hindering the platform's usability, security, and scalability. Traditional metaverse e-commerce platforms typically utilize centralized data storage and plaintext processing, making user behavior data, social relationships, and transaction information vulnerable to theft or tampering by malicious attackers. Although some platforms have attempted to incorporate blockchain technology to ensure the immutability of transaction records, the lack of encryption for raw data means that sensitive information such as user browsing preferences and 3D product features can still be reverse engineered through data correlation analysis. Furthermore, frequent data decryption during transaction verification further increases the risk of privacy breaches. Existing deep learning models struggle to effectively capture key features when processing multimodal data (such as 3D models, textures, and materials). Traditional retrieval models, which often rely on Euclidean distance or cosine similarity for matching, are unable to distinguish between local details in complex 3D structures, resulting in low product recommendation accuracy. At the same time, the model training process fails to incorporate optimization strategies for multimodal data, such as the attention mechanism, further limiting the efficiency of feature parsing. Metaverse e-commerce platforms need to support massive numbers of users simultaneously rendering virtual goods and interacting with them in real time. Existing solutions often rely on centralized cloud-based computing, resulting in long data transmission paths and significant rendering delays. While edge computing technology has been partially adopted, it's not combined with lightweight model deployment and dynamic resource scheduling strategies. The imbalanced computing load on edge nodes is prominent, making it difficult to meet low-latency interaction requirements. The platform needs to synchronize user data, model parameters, and transaction records between the cloud and edge nodes. Traditional blockchain technology suffers from inefficient synchronization due to the large amount of on-chain data and complex consensus mechanisms. Furthermore, model updates often utilize a centralized training model that requires uploading raw data to the cloud, posing a risk of privacy leakage. Furthermore, there's a lack of efficient framework support for cross-node collaborative training. Summary of the Invention

[0003] The present invention provides a method for blockchain encryption and deep learning retrieval for a metaverse e-commerce platform, which can solve the above-mentioned problems.

[0004] In order to solve the above problems, the technical solutions adopted by the present invention are as follows:

[0005] A method for blockchain encryption and deep learning retrieval for a metaverse e-commerce platform, comprising: using homomorphic encryption technology to protect the privacy of user behavior data, virtual product features, and social network relationships, and combining it with a zero-knowledge proof protocol to generate tamper-proof transaction credentials to construct an encrypted data set containing multimodal features; designing a deep learning retrieval model, enhancing the ability to parse 3D product features by introducing an attention mechanism, and optimizing model parameters using a contrast loss function; obtaining a lightweight retrieval model by performing quantitative pruning on the deep learning retrieval model, and deploying the lightweight retrieval model to distributed edge nodes, combining it with a dynamic resource scheduling algorithm to achieve real-time rendering and low-latency interaction of virtual products; on the edge nodes, introducing a federated learning framework to incrementally update the lightweight retrieval model, and achieving continuous optimization of the model while ensuring data privacy through local model training and global model aggregation; establishing a cloud-edge collaborative blockchain synchronization mechanism, using sharding technology to reduce the amount of on-chain data transmission, and ensuring data consistency between the cloud and edge nodes through a consensus algorithm.

[0006] Specifically, the use of homomorphic encryption technology to protect the privacy of user behavior data, virtual product characteristics, and social network relationships includes:

[0007] Obtaining raw data on user behavior, virtual product characteristics, and social network relationships;

[0008] Inputting the original data into a homomorphic encryption module to generate an encrypted data set;

[0009] Combined with the zero-knowledge proof protocol, an unalterable transaction certificate is generated and stored in the blockchain.

[0010] Specifically, the process of designing the deep learning retrieval model includes:

[0011] Constructing a basic deep learning model, wherein the basic deep learning model includes an input layer, a hidden layer, and an output layer;

[0012] Introducing an attention mechanism module into the hidden layer to enhance the ability to analyze 3D product features;

[0013] Define the contrastive loss function for optimizing model parameters.

[0014] Specifically, the lightweight retrieval model is obtained by performing quantization pruning processing on the deep learning retrieval model, including:

[0015] Obtain the trained deep learning retrieval model;

[0016] Model pruning technology is used to remove redundant parameters, and dynamic range quantization method is used to quantize floating-point weights into fixed-bit-width integers;

[0017] Perform performance tests on the quantized model and adjust pruning and quantization parameters based on the test results to ensure the stability of the model performance.

[0018] Specifically, the dynamic resource scheduling algorithm includes:

[0019] Obtain real-time load information of edge nodes, including computing demand, network latency, and node utilization;

[0020] Calculating resource allocation priorities based on the real-time load information;

[0021] According to the resource allocation priority, the task allocation strategy of the edge node is dynamically adjusted.

[0022] Specifically, the introduction of the federated learning framework to incrementally update the lightweight retrieval model includes:

[0023] On edge nodes, the lightweight retrieval model is locally trained using local data.

[0024] Upload the trained local model parameters to the cloud;

[0025] In the cloud, local model parameters uploaded by multiple edge nodes are aggregated to generate a global model;

[0026] The global model is sent to each edge node to complete the model update.

[0027] Specifically, the establishment of a cloud-edge collaborative blockchain synchronization mechanism includes:

[0028] Divide the on-chain data into multiple shards, each containing part of the data;

[0029] Calculate the data transmission time after sharding;

[0030] The consensus algorithm is used to verify the shard data to ensure data consistency between the cloud and edge nodes.

[0031] Specifically, the real-time rendering and low-latency interaction of virtual goods include:

[0032] Obtain virtual product data requested by the user;

[0033] Inputting the virtual commodity data into a lightweight retrieval model to generate a feature vector required for rendering;

[0034] Based on the feature vector, calling the rendering engine of the edge node to perform real-time rendering;

[0035] The rendering results are returned to the user terminal to complete low-latency interaction.

[0036] Specifically, the zero-knowledge proof protocol generates an unalterable transaction voucher, including:

[0037] Obtain transaction data and perform hash operations on it to generate a transaction summary;

[0038] Using a zero-knowledge proof algorithm to verify the transaction summary and generate a transaction voucher;

[0039] The transaction voucher is stored in the blockchain to ensure that the transaction data cannot be tampered with.

[0040] Specifically, constructing an encrypted dataset containing multimodal features includes:

[0041] Obtaining raw data on user behavior, virtual product characteristics, and social network relationships;

[0042] Preprocessing the raw data, including data cleaning, feature extraction, and normalization operations;

[0043] The preprocessed data is input into the homomorphic encryption module to generate an encrypted multimodal feature dataset.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] Through homomorphic encryption and zero-knowledge proof protocols, user data can be encrypted and stored, and transaction credentials cannot be tampered with, effectively preventing privacy leaks and transaction fraud; the deep learning model with attention mechanism combined with contrast loss function optimization significantly improves the analysis accuracy and retrieval efficiency of multimodal 3D product features; lightweight models are deployed on edge nodes, combined with dynamic resource allocation strategies to reduce computing load and achieve real-time rendering and low-latency interaction of virtual products; the federated learning framework supports local training and global aggregation of edge nodes, and realizes model iterative updates without sharing original data; cloud-edge collaborative blockchain sharding technology reduces the amount of on-chain data transmission, and combined with the consensus algorithm to ensure the efficiency and consistency of cross-node data synchronization.

[0046] In order to make the above-mentioned objects, features and advantages of the present invention more clearly understood, embodiments of the present invention are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0048] Figure 1 This is a flowchart of the method of the present invention, which includes the encryption of user behavior data, the generation of transaction credentials using the zero-knowledge proof protocol, and the construction process of the multimodal feature encryption data set.

[0049] Figure 2 This is a schematic diagram of the blockchain synchronization mechanism for cloud-edge collaboration, showing the application of sharding technology in on-chain data transmission, the incremental update process of the federated learning framework, and the operating logic of the dynamic resource scheduling algorithm. DETAILED DESCRIPTION

[0050] The present invention provides a method for blockchain encryption and deep learning retrieval for the Metaverse e-commerce platform. Figure 1 and Figure 2 This paper describes its specific implementation in detail. Taking the actual application scenario of a large-scale metaverse e-commerce company as an example, this paper demonstrates how to achieve user privacy protection, accurate virtual product recommendations, and high-concurrency, low-latency interaction through homomorphic encryption technology, zero-knowledge proof protocols, improved deep learning retrieval models, quantitative pruning processing, dynamic resource scheduling algorithms, federated learning frameworks, and cloud-edge collaborative blockchain synchronization mechanisms.

[0051] Construct an encrypted dataset containing multimodal features:

[0052] Obtain raw data on user behavior, virtual product characteristics, and social network relationships. In practice, this data comes from user operations on the Metaverse platform, including clicks, browsing, and purchases, as well as feature data such as virtual product 3D models, material parameters, and texture information, as well as social interaction records between users.

[0053] Preprocess the raw data, including data cleaning, feature extraction, and normalization. For example, for user behavior data, invalid or duplicate operation records are removed, and discrete features are converted to numerical form through one-hot encoding. For virtual product features, the key point coordinates, normal vector distribution, and texture mapping matrix of the 3D model are extracted. For social network relationships, the adjacency matrix is ​​used to represent the connection strength between users.

[0054] The preprocessed data is fed into the homomorphic encryption module to generate an encrypted multimodal feature dataset. This module uses the Paillier encryption algorithm, allowing direct addition and multiplication operations on encrypted data without leaking the original data, thus avoiding the risk of privacy leaks during the decryption process. Furthermore, it integrates a zero-knowledge proof protocol to generate tamper-proof transaction vouchers.

[0055] Get the transaction data and perform a hash operation on it to generate a transaction summary.

[0056] A zero-knowledge proof algorithm is used to verify the transaction summary and generate a transaction voucher. Specifically, the transaction data includes product order information, payment amount, and timestamp. A fixed-length transaction digest is generated using the SHA-256 algorithm. The authenticity of the transaction is then verified using a zero-knowledge proof algorithm, generating a transaction voucher that only contains the proof information without revealing the specific content.

[0057] Storing transaction credentials in the blockchain ensures that transaction data cannot be tampered with. Figure 1 As shown in the figure, this module implements the complete process from original data to encrypted data and then to tamper-proof transaction credentials, providing a secure and reliable data foundation for subsequent modules.

[0058] Deep learning retrieval model design and optimization:

[0059] Build a basic deep learning model consisting of an input layer, hidden layers, and an output layer. The input layer receives an encrypted multimodal feature dataset, the hidden layer is composed of a multi-layer fully connected neural network, and the output layer is responsible for generating the final feature vector.

[0060] An attention mechanism module is introduced into the hidden layer to enhance the ability to capture key features of 3D products. Specifically, the attention mechanism module calculates the importance weight of each feature point and prioritizes key areas of the 3D product, such as shape outline, material reflectivity, and texture details.

[0061] Define the contrast loss function to optimize the model parameters. The formula is:

[0062]

[0063] in, represents the feature vector extracted by the deep learning model, is the Euclidean distance function, A label indicating whether a pair of samples is similar. m is a preset boundary value used to distinguish the distance between positive and negative sample pairs. During training, by minimizing the contrastive loss function, the feature vectors of similar samples are brought closer together, while the feature vectors of dissimilar samples are further apart, significantly improving retrieval accuracy and generalization capabilities.

[0064] Quantization pruning of lightweight retrieval models:

[0065] Get the trained deep learning retrieval model.

[0066] Model pruning techniques are used to remove redundant parameters, and dynamic range quantization (DRS) is used to quantize floating-point weights into fixed-bit-width integers. Model pruning analyzes the importance of weights and removes parameters that contribute little to model performance, such as weights below a certain threshold. Dynamic range quantization uses statistical analysis of weight distribution to determine the quantization interval and bit width, such as quantizing 32-bit floating-point numbers into 8-bit integers, thereby reducing the model's computational complexity and storage requirements.

[0067] The quantized model is performance-tested, and pruning and quantization parameters are adjusted based on the test results to ensure model performance stability. For example, if the model accuracy drops by more than 5% during testing, the pruning ratio is appropriately reduced or the quantization bit width is increased until performance requirements are met. The resulting lightweight retrieval model is suitable for deployment on resource-constrained edge nodes, significantly reducing computational overhead and storage pressure.

[0068] Edge node deployment and dynamic resource scheduling:

[0069] Obtain the virtual product data requested by the user. For example, a user requests to view the rendering of a 3D furniture model through a terminal device.

[0070] Virtual product data is fed into a lightweight retrieval model to generate the feature vectors required for rendering. The feature vectors contain key information such as the product's shape, material, and texture, providing a basis for subsequent rendering.

[0071] Based on the feature vector, the edge node's rendering engine is called for real-time rendering. The edge node uses the GPU to accelerate the rendering engine and quickly generate high-quality 3D images.

[0072] The rendering results are returned to the user terminal, completing low-latency interaction. Since the edge node is close to the user terminal, the data transmission path is significantly shortened, and the delay of rendering and interaction is effectively controlled.

[0073] The design process of the dynamic resource scheduling algorithm is as follows:

[0074] Obtain real-time load information for edge nodes, including computing demand, network latency, and node utilization. For example, computing demand refers to the CPU and GPU resources required for the current task, network latency refers to the transmission time of data packets in the network, and node utilization refers to the proportion of edge node resources occupied. Based on real-time load information, calculate the resource allocation priority using the formula:

[0075]

[0076] in, Indicates the resource allocation priority of edge nodes, is the computational requirement of the current task, For network delay, is the node utilization, is the weight coefficient used to balance various indicators. For example, when the network delay is high, increase The value of is used to prioritize resource allocation and reduce latency.

[0077] According to the resource allocation priority, the task allocation strategy of the edge node is dynamically adjusted. For example, when the load of an edge node is too high, some of its tasks will be migrated to the node with lower load, so as to achieve efficient resource allocation. Figure 2 As shown in the figure, this module significantly improves the overall system efficiency and user experience through the feedback of real-time load information.

[0078] Implementation of the Federated Learning Framework and Cloud-Edge Collaborative Blockchain Synchronization Mechanism:

[0079] On edge nodes, lightweight retrieval models are trained locally using local data. For example, edge nodes collect user behavior data and update model parameters without uploading the original data, thus ensuring data privacy.

[0080] The trained local model parameters are uploaded to the cloud. The cloud receives the local model parameters uploaded by multiple edge nodes and aggregates them.

[0081] In the cloud, local model parameters uploaded by multiple edge nodes are aggregated to generate a global model. For example, the FedAvg algorithm is used to calculate the weighted average of the local model parameters to generate a new global model.

[0082] Send the global model to each edge node to complete the model update.

[0083] Through the above steps, the federated learning framework achieves continuous optimization of the model while ensuring data privacy, solving the privacy leakage problem in the traditional centralized training model.

[0084] The design process of the cloud-edge collaborative blockchain synchronization mechanism is as follows:

[0085] Divide the on-chain data into multiple shards, each containing a portion of the data. For example, divide transaction data into several shards based on timestamps, with each shard containing transaction records within a certain time period.

[0086] Calculate the data transmission time after sharding. The formula is:

[0087]

[0088] in, Indicates the data transmission time after sharding, N is the total data volume, is the number of shards, and P is the number of nodes participating in the consensus.

[0089] The PBFT consensus algorithm is used to verify the shard data to ensure data consistency between the cloud and edge nodes. For example, the PBFT algorithm verifies the correctness of the shard data through a multi-round voting mechanism and synchronizes it to all nodes. Figure 2 As shown in the figure, this module effectively reduces the communication overhead of blockchain synchronization through sharding technology and consensus algorithm, and improves the scalability and data consistency of the system.

[0090] In summary, the technical solution of the present invention achieves the unification of user privacy protection, accurate recommendation of virtual goods and high concurrency and low latency interaction through the synergy of the above modules. Specifically, homomorphic encryption technology and zero-knowledge proof protocol ensure the security of user behavior data and transaction data; the improved deep learning retrieval model significantly improves the parsing ability and retrieval accuracy of the multimodal features of 3D goods through the attention mechanism and contrast loss function optimization; the quantitative pruning process and dynamic resource scheduling algorithm reduce the computational complexity and storage requirements of the model, while improving the overall efficiency of the system; the federated learning framework and the cloud-edge collaborative blockchain synchronization mechanism achieve continuous optimization of the model and data consistency while ensuring data privacy. The technical solution of the present invention has the following technical effects: through the application of homomorphic encryption technology and zero-knowledge proof protocol, the problem of user privacy leakage is solved; through the design and optimization of the improved deep learning retrieval model, the recommendation accuracy of virtual goods is improved; through the quantitative pruning process of the lightweight retrieval model and the distributed deployment of edge nodes, real-time rendering and low-latency interaction of virtual goods are achieved; through the design of the federated learning framework and the cloud-edge collaborative blockchain synchronization mechanism, continuous optimization of the model and data consistency are achieved. In particular, the technical solution of the present invention provides strong technical support for the sustainable development of metaverse e-commerce.

[0091] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A method for blockchain encryption and deep learning retrieval for a Metaverse e-commerce platform, characterized in that: include: Homomorphic encryption technology is used to protect the privacy of user behavior data, virtual product characteristics, and social network relationships. In combination with the zero-knowledge proof protocol, tamper-proof transaction vouchers are generated to construct an encrypted dataset containing multimodal features. Designing a deep learning retrieval model, which specifically includes: building a deep learning model consisting of an input layer, a hidden layer, and an output layer, where the output layer is responsible for generating the final feature vector; An attention mechanism module is introduced into the hidden layer to enhance the ability to capture key features of 3D products by calculating the importance weights of feature points. A contrastive loss function is defined, which combines a loss term proportional to the distance between the feature vectors of similar sample pairs and a loss term that generates a penalty only when the distance between the feature vectors of dissimilar sample pairs is less than a preset distance threshold. The model parameters are optimized by minimizing the contrastive loss function, thereby bringing the feature vectors of similar samples closer together and the feature vectors of dissimilar samples further apart. By performing quantitative pruning on the deep learning retrieval model, a lightweight retrieval model is obtained, and the lightweight retrieval model is deployed to distributed edge nodes, and combined with a dynamic resource scheduling algorithm, real-time rendering and low-latency interaction of virtual goods are realized; on the edge nodes, a federated learning framework is introduced to incrementally update the lightweight retrieval model, and through local model training and global model aggregation, continuous optimization of the model is achieved while ensuring data privacy; a cloud-edge collaborative blockchain synchronization mechanism is established, sharding technology is used to reduce the amount of on-chain data transmission, and a consensus algorithm is used to ensure data consistency between the cloud and edge nodes.

2. The method for blockchain encryption and deep learning retrieval for the Metaverse e-commerce platform according to claim 1, characterized in that: The use of homomorphic encryption technology to protect the privacy of user behavior data, virtual product characteristics, and social network relationships includes: Obtaining raw data on user behavior, virtual product characteristics, and social network relationships; Inputting the original data into a homomorphic encryption module to generate an encrypted data set; Combined with the zero-knowledge proof protocol, an unalterable transaction certificate is generated and stored in the blockchain.

3. The method for blockchain encryption and deep learning retrieval for the Metaverse e-commerce platform according to claim 1 is characterized in that: The lightweight retrieval model is obtained by performing quantization pruning processing on the deep learning retrieval model, including: Obtain the trained deep learning retrieval model; Model pruning technology is used to remove redundant parameters, and dynamic range quantization method is used to quantize floating-point weights into fixed-bit-width integers; Perform performance tests on the quantized model and adjust pruning and quantization parameters based on the test results to ensure the stability of the model performance.

4. The method for blockchain encryption and deep learning retrieval for the Metaverse e-commerce platform according to claim 1, characterized in that: The dynamic resource scheduling algorithm includes: Obtain real-time load information of edge nodes, including computing demand, network latency, and node utilization; Calculating resource allocation priorities based on the real-time load information; According to the resource allocation priority, the task allocation strategy of the edge node is dynamically adjusted.

5. The method for blockchain encryption and deep learning retrieval for the Metaverse e-commerce platform according to claim 1 is characterized in that: The federated learning framework is introduced to incrementally update the lightweight retrieval model, including: On edge nodes, the lightweight retrieval model is locally trained using local data. Upload the trained local model parameters to the cloud; In the cloud, local model parameters uploaded by multiple edge nodes are aggregated to generate a global model; The global model is sent to each edge node to complete the model update.

6. The method for blockchain encryption and deep learning retrieval for the Metaverse e-commerce platform according to claim 1, characterized in that: The blockchain synchronization mechanism for establishing cloud-edge collaboration includes: Divide the on-chain data into multiple shards, each containing part of the data; Calculate the data transmission time after sharding; The consensus algorithm is used to verify the shard data to ensure data consistency between the cloud and edge nodes.

7. The method for blockchain encryption and deep learning retrieval for the Metaverse e-commerce platform according to claim 1, characterized in that: The real-time rendering and low-latency interaction of the virtual goods include: Obtain virtual product data requested by the user; Inputting the virtual commodity data into a lightweight retrieval model to generate a feature vector required for rendering; Based on the feature vector, calling the rendering engine of the edge node to perform real-time rendering; The rendering results are returned to the user terminal to complete low-latency interaction.

8. The method for blockchain encryption and deep learning retrieval for the Metaverse e-commerce platform according to claim 1, characterized in that: The zero-knowledge proof protocol generates an unalterable transaction certificate, including: Obtain transaction data and perform hash operations on it to generate a transaction summary; Using a zero-knowledge proof algorithm to verify the transaction summary and generate a transaction voucher; The transaction voucher is stored in the blockchain to ensure that the transaction data cannot be tampered with.

9. The method for blockchain encryption and deep learning retrieval for the Metaverse e-commerce platform according to claim 1, characterized in that: The construction of an encrypted dataset containing multimodal features includes: Obtaining raw data on user behavior, virtual product characteristics, and social network relationships; Preprocessing the raw data, including data cleaning, feature extraction, and normalization operations; The preprocessed data is input into the homomorphic encryption module to generate an encrypted multimodal feature dataset.

Images