Intelligent recommendation system for big data and personalized pushing method thereof

Abstract

This invention discloses an intelligent recommendation system and personalized push method for big data, relating to the field of big data processing technology. The system includes acquiring local federated state information from user terminals; generating an integrated collaborative instruction based on the local federated state information; the user terminal processing and adding noise to local user data according to the integrated collaborative instruction; and a cloud server receiving differentiated data from various user terminals and updating the global federated recommendation model. This invention dynamically adapts to data processing methods and privacy budget allocation, systematically solving the balance problem between privacy protection, resource consumption, and model quality in federated recommendation. It significantly improves the system's robustness and overall performance in complex environments, achieving system self-evolution and continuous improvement in recommendation accuracy, thus providing a secure, efficient, and adaptive integrated solution for personalized services in big data environments.

Description

Technical Field

[0001] This invention relates to the field of big data processing technology, and in particular to an intelligent recommendation system for big data and its personalized push method. Background Technology

[0002] With the booming development of mobile internet and the Internet of Things, data volume is exploding. Big data-based intelligent recommendation systems have become core services in many fields such as e-commerce, content platforms, and social networks. Traditional centralized recommendation systems require aggregating massive amounts of user behavior data to a cloud-based central server for unified modeling and analysis. This approach increasingly faces two major challenges: First, user data privacy and compliance risks are becoming increasingly prominent, with strict data protection regulations in various countries (such as GDPR and CCPA) restricting the free aggregation of data. Second, data generated by user terminal devices is heterogeneous, massive, and real-time; uploading it entirely to the cloud for processing would bring enormous network bandwidth pressure, server computing load, and user-side communication energy consumption. To address these privacy challenges, federated learning, as a distributed machine learning paradigm, should... Born from this, it allows models to be trained locally on terminal devices. However, when deploying federated recommendation systems for big data, existing technical solutions still face the following bottlenecks that urgently need to be addressed: the inherent trade-off between privacy protection and model performance: existing solutions typically employ fixed differential privacy mechanisms, i.e., adding noise of fixed intensity to the terminal or cloud; the contradiction between heterogeneous terminal resources and a single processing mode: user terminals vary greatly in terms of computing power, battery life, and network conditions. Existing federated recommendation solutions often stipulate a unified local processing mode (such as fixed-length gradient calculation), ignoring this heterogeneity; and the mismatch between static strategies and dynamic environments: most existing data upload and processing strategies are static or semi-static, unable to respond in real time to dynamic environmental factors such as network congestion, battery changes, and users' immediate adjustments to privacy preferences.

[0003] However, current common solutions have many drawbacks, including: existing technologies adopt static privacy protection strategies and single data processing modes, which cannot be dynamically adapted according to the heterogeneous resource status, network conditions, data value, and user preferences of terminals, making it difficult to achieve an optimal balance between privacy, efficiency, and model accuracy; the cloud aggregation mechanism is rigid and cannot effectively accommodate and integrate the differentiated data forms generated by terminal adaptation and optimization, limiting the overall flexibility of the system; the various links are isolated from each other, lacking closed-loop optimization capabilities based on effect feedback, and privacy protection has failed to form a collaborative resource allocation mechanism between the terminal and the cloud, making it difficult to simultaneously ensure high standards of privacy security, resource efficiency, and recommendation quality in a complex and ever-changing big data environment. Summary of the Invention

[0004] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0005] In view of the problems existing in the current intelligent recommendation system and personalized push method for big data, this invention is proposed.

[0006] Therefore, the purpose of this invention is to provide an intelligent recommendation system and its personalized push method for big data. This system addresses the shortcomings of existing technologies that employ static privacy protection strategies and single data processing models. These systems cannot dynamically adapt to the heterogeneous resource status, network conditions, data value, and user preferences of terminals, leading to difficulties in achieving an optimal balance between privacy, efficiency, and model accuracy. Furthermore, the rigid cloud aggregation mechanism cannot effectively accommodate and integrate the differentiated data formats generated by terminal-side adaptation optimization, limiting the overall flexibility of the system. The fragmented nature of each component lacks closed-loop optimization capabilities based on feedback, and privacy protection fails to establish a collaborative resource allocation mechanism between the terminal and cloud. Consequently, it is difficult to simultaneously guarantee high standards of privacy security, resource efficiency, and recommendation quality in a complex and ever-changing big data environment.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, embodiments of the present invention provide an intelligent recommendation and personalized push method for big data, which includes obtaining local federated state information of user terminals; generating an integrated collaborative instruction based on the local federated state information, the instruction simultaneously determining the data processing form and privacy budget allocation scheme; the user terminal processing and adding noise to local user data according to the integrated collaborative instruction, and uploading the results; and a cloud server receiving differentiated data from various user terminals, performing secure aggregation, and updating the global federated recommendation model.

[0008] As a preferred embodiment of the intelligent recommendation and personalized push method for big data described in this invention, the information includes computing resource status, network status, data value assessment of user data to be uploaded, and user privacy preferences; the data processing method specifies uploading one of the following to the cloud: raw data, locally generated embedded vectors, or locally encrypted data; the privacy budget allocation scheme dynamically allocates the budget ratio for adding differential privacy noise on the terminal side and the cloud side within a preset total privacy budget.

[0009] As a preferred embodiment of the intelligent recommendation and personalized push method for big data described in this invention, the acquisition of the local joint state information of the user terminal specifically includes: real-time collection of the user terminal's computing resource status, network connection status, data value attributes of the user behavior data to be uploaded, and the privacy preference level actively set by the user; the computing resource status includes at least the currently available computing power and remaining battery power; the network connection status includes at least the uplink bandwidth and transmission latency; the data value attributes are determined by analyzing the novelty, sparsity, and potential contribution of the user behavior data to the global model.

[0010] As a preferred embodiment of the intelligent recommendation and personalized push method for big data described in this invention, the method comprises: generating an integrated collaborative instruction based on the local joint state information, which simultaneously determines the data processing form and the privacy budget allocation scheme, specifically as follows: inputting the computing resource status, network connection status, data value attributes, and user privacy preference level into a lightweight decision model; the lightweight decision model simultaneously outputting the selection result of the data processing form and the privacy budget allocation scheme; the selection of the data processing form is based on a comprehensive trade-off between data upload timeliness, computing overhead, and privacy exposure risks; the privacy budget allocation scheme is dynamically generated based on a comprehensive assessment of the local resource consumption of adding noise on the terminal side and the communication privacy risks of adding noise on the cloud side.

[0011] As a preferred embodiment of the intelligent recommendation and personalized push method for big data described in this invention, the user terminal processes local user data according to an integrated collaborative instruction, adds noise, and uploads the results. Specifically, according to the data processing format specified in the integrated collaborative instruction, the local edge computing module is invoked to perform corresponding operations: when the format is uploading raw data, the raw user behavior sequence to be uploaded is directly cached; when the format is uploading locally generated embedding vectors, the raw user behavior sequence is encoded into short-term interest embedding vectors using a local lightweight neural network model; when the format is uploading locally encrypted data, the raw user behavior... The sequence or generated embedding vector is preprocessed with homomorphic encryption or secure multi-party computation; according to the privacy budget allocation scheme specified in the integrated collaborative instruction, differential privacy noise that meets the budget requirements is injected into the corresponding data processing stage: if the budget allocation scheme indicates that noise is added on the terminal side, then Laplace noise or Gaussian noise of appropriate intensity is added to the cached, encoded or encrypted data; if the budget allocation scheme indicates that all or part of the noise addition is delayed to the cloud, then weak noise or only identifiers are added to the data to be uploaded; the data after morphological processing and noise injection is encapsulated into data packets according to a preset secure communication protocol and uploaded to the cloud server via network connection.

[0012] As a preferred embodiment of the intelligent recommendation and personalized push method for big data described in this invention, the cloud server receives differentiated data from various user terminals, performs secure aggregation, and updates the global federated recommendation model. Specifically, it receives data packets with different data processing formats from different user terminals in parallel and decodes and verifies these data packets. For data packets identified as raw data, user behavior sequences are directly extracted. For data packets identified as embedded vectors, noisy short-term interest embedding vectors are extracted. For data packets identified as locally encrypted data, secure decoding is performed using the corresponding key or protocol. Data with added differential privacy noise is directly used for aggregation. For data where insufficient noise or only identifiers are added on the terminal side, differential privacy noise of appropriate intensity is injected on the cloud side according to its remaining privacy budget. Dynamic aggregation weights are calculated for each user's contribution data based on the data value attributes, privacy budget consumption, and data quality confidence of the data uploaded by each user terminal. These dynamic aggregation weights are then used to perform a weighted average of all privacy-post-processed user data or embedding vectors to generate the current round's aggregation update amount. This aggregation update amount is then integrated into the parameters of the current global federated recommendation model through secure multi-party computation or homomorphic encryption to complete the model update iteration. Finally, some metadata of the updated global model or the initial parameters required for the next round of training are securely distributed to the participating user terminals.

[0013] As a preferred embodiment of the intelligent recommendation and personalized push method for big data described in this invention, the method includes: generating and pushing personalized recommendation content to the corresponding user terminal based on the updated global federated recommendation model; dynamically adjusting the data value attribute calculated for the terminal in subsequent rounds based on the user terminal's feedback and interaction data on the pushed content, and using the adjusted data value attribute in the local federated state information of subsequent rounds.

[0014] Secondly, to further address the aforementioned technical problems, the present invention provides an intelligent recommendation system for big data, comprising: an information acquisition module for acquiring local federated state information of user terminals; an instruction generation module for generating an integrated collaborative instruction based on the local federated state information, wherein the instruction simultaneously determines the data processing format and privacy budget allocation scheme; a data upload module for user terminals to process local user data in a corresponding format and add noise according to the integrated collaborative instruction, and upload the results; and a model update module for a cloud server to receive differentiated data from various user terminals, perform secure aggregation, and update the global federated recommendation model.

[0015] Thirdly, embodiments of the present invention provide a computer device, including a memory and a processor, wherein the memory stores a computer program, and the computer program, when executed by the processor, implements any step of the intelligent recommendation and personalized push method for big data as described in the first aspect of the present invention.

[0016] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the intelligent recommendation and personalized push method for big data as described in the first aspect of the present invention.

[0017] The beneficial effects of this invention are as follows: By introducing an integrated collaborative decision-making mechanism based on real-time terminal status, this invention dynamically adapts to data processing patterns and privacy budget allocation, systematically solving the balance problem between privacy protection, resource consumption, and model quality in federated recommendations; it allows heterogeneous terminals to participate and contribute value in a differentiated manner, constructing a flexible privacy protection chain for end-to-cloud collaboration, significantly improving the robustness and overall efficiency of the system in complex environments; through closed-loop optimization of core parameters driven by user feedback, it achieves the self-evolution of the system and continuous improvement of recommendation accuracy, thus providing a safe, efficient, and adaptive integrated solution for personalized services in a big data environment. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a flowchart illustrating the implementation of the present invention in Example 1.

[0019] Figure 2 This is a dynamic adjustment diagram of the present invention in Example 1. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0023] Example 1 Reference Figure 1 and Figure 2 This is the first embodiment of the present invention, which provides an intelligent recommendation and personalized push method for big data, including the following steps: S1: Obtain the local union state information of the user terminal.

[0024] Furthermore, the information includes computing resource status, network status, data value assessment of user data to be uploaded, and user privacy preferences.

[0025] Preferably, the local federation status information of the user terminal is obtained, the specific content of which is as follows: It collects in real time the computing resource status of user terminals, network connection status, data value attributes of user behavior data to be uploaded, and privacy preference levels actively set by users.

[0026] The status of computing resources includes at least the currently available computing power and the remaining power.

[0027] Network connectivity status includes at least uplink bandwidth and transmission latency.

[0028] The value attributes of data are determined by analyzing the novelty, sparsity, and potential contribution of user behavior data to the global model. The specific formula is as follows: ; In the formula, For data value attributes; The subset of user behavior data to be evaluated; The score represents the novelty of the data; The sparsity score of the data; The expected utility of the data for the current global model; , , The normalized weights for the sub-dimensions.

[0029] For example, consider a user using an e-commerce application. Their phone currently has 65% battery remaining, moderate available CPU computing power, and is connected to a Wi-Fi network with an uplink bandwidth of 5Mbps and a latency of 50ms. The system detects that the user has just generated a sequence of browsing and saving multiple newly launched niche brand products. Through calculation, the system determines that this behavioral sequence has high novelty and sparsity, and the user previously set their privacy preference to "medium protection". Finally, the system integrates these real-time collected computing power, battery level, network indicators, calculated high data value attributes, and the user's privacy level to form a structured local joint state information.

[0030] S2: Based on local joint state information, generate an integrated collaborative instruction that simultaneously determines the data processing format and privacy budget allocation scheme.

[0031] Specifically, the data processing mode specifies uploading one of the following to the cloud: raw data, locally generated embedded vectors, or locally encrypted data.

[0032] Furthermore, the privacy budget allocation scheme dynamically allocates the budget ratio for adding differential privacy noise on the terminal side and the cloud side within the preset total privacy budget.

[0033] Preferably, based on local joint state information, an integrated collaborative instruction is generated. This instruction simultaneously determines the data processing format and privacy budget allocation scheme, as detailed below: Input computing resource status, network connectivity status, data value attributes, and user privacy preference levels into a lightweight decision model.

[0034] The lightweight decision model simultaneously outputs the selection results of data processing mode and privacy budget allocation scheme.

[0035] The choice of data processing method is based on a comprehensive trade-off between data upload timeliness, computational overhead, and privacy exposure risks.

[0036] The privacy budget allocation scheme is dynamically generated based on a comprehensive assessment of the local resource consumption of adding noise on the terminal side and the communication privacy risks of adding noise on the cloud side.

[0037] Preferably, this step constructs a multi-dimensional state space that integrates computing resources, network status, data value, and privacy preferences. It then uses a lightweight decision model to generate integrated instructions that simultaneously determine the data processing format and privacy budget allocation. This achieves a paradigm shift from static trade-offs to dynamic optimization. It can dynamically approximate the Pareto optimal solution of the system in terms of privacy protection, resource consumption, and model performance based on the real-time environment. Furthermore, by explicitly evaluating and incorporating data value into decisions, it intelligently guides heterogeneous terminals to make high-quality contributions in a way that best suits their own state. This solves the problems of unfair participation and overall performance bottlenecks caused by differences in terminal resources at the system level.

[0038] For example, the system inputs the collected joint state information into a deployed lightweight neural network decision model. After comprehensively evaluating multiple factors such as "high data value", "average network bandwidth", "moderate user privacy requirements" and "sufficient terminal power and computing power", the model outputs a decision: selects "upload locally generated embedding vectors" as the data processing form, and dynamically allocates the total privacy budget for this task to 70% for adding noise on the terminal side and 30% for the cloud side, thereby optimizing network transmission load while ensuring data utility and privacy.

[0039] S3: The user terminal processes the local user data according to the integrated collaborative instructions, adds noise, and uploads the results.

[0040] Preferably, the user terminal processes the local user data and adds noise according to the integrated collaborative instructions, and then uploads the results, as follows: Based on the data processing format specified in the integrated collaborative instruction, the local edge computing module is invoked to perform the corresponding operation: When the data is in the form of uploading raw data, the original user behavior sequence to be uploaded is directly cached.

[0041] When the form is an uploaded locally generated embedding vector, the original user behavior sequence is encoded into a short-term interest embedding vector using a local lightweight neural network model.

[0042] When the data is uploaded locally encrypted, the original user behavior sequence or the generated embedding vector is preprocessed with homomorphic encryption or secure multi-party computation.

[0043] Based on the privacy budget allocation scheme specified in the integrated collaboration instructions, differential privacy noise that meets the budget requirements is injected into the corresponding data processing stage: If the budget allocation scheme indicates that noise should be added on the terminal side, then Laplace noise or Gaussian noise of appropriate intensity should be added to the cached, encoded, or encrypted data.

[0044] If the budget allocation scheme indicates that noise additions are all or partly delayed to the cloud, then weak noise or only identifiers are added to the data to be uploaded.

[0045] The data, after morphological processing and noise injection, is encapsulated into data packets according to a preset secure communication protocol and uploaded to the cloud server via a network connection.

[0046] Preferably, by treating the privacy budget as a resource that can be flexibly partitioned and allocated between the edge and the cloud, and instructing the terminal to execute differentiated data processing forms (raw data / embedded vectors / encrypted data) and staged noise injection based on integrated decision-making, the "spatial elasticity" of the privacy protection mechanism is achieved. By coordinating the optimization of the computing overhead on the terminal side and the communication privacy risks on the cloud side, the overall marginal cost of privacy protection is minimized. At the same time, this step supports the construction of a computing pipeline that is dynamically assembled according to the state, so that the data processing flow can adapt to the network and terminal load, providing an efficient and feasible technical path for implementing high-intensity privacy protection in resource-constrained dynamic edge environments.

[0047] For example, after receiving the above-mentioned integrated instruction, the user terminal first calls the local lightweight encoder to convert the user behavior sequence into a 256-dimensional short-term interest embedding vector; then, according to the instruction's requirement of "70% of the budget is added on the terminal side", an appropriate amount of Gaussian noise is added to the embedding vector; next, the noisy embedding vector is encapsulated into a data packet and uploaded to the cloud server through a TLS secure channel, thereby completing a data reporting process that optimizes resources, privacy, and efficiency.

[0048] S4: The cloud server receives differentiated data from various user terminals, performs secure aggregation, and updates the global federated recommendation model.

[0049] Preferably, the cloud server receives differentiated data from various user terminals, performs secure aggregation, and updates the global federated recommendation model, as detailed below: It receives data packets with different data processing formats from different user terminals in parallel, and decodes and verifies the data packets.

[0050] For data packets identified as raw data formats, user behavior sequences are extracted directly.

[0051] For data packets identified as embedding vectors, extract noisy short-term interest embedding vectors.

[0052] For data packets identified as locally encrypted data, secure decoding is performed using the corresponding key or protocol.

[0053] Data with fully added differential privacy noise is used directly for aggregation.

[0054] For data that has not had sufficient noise added on the terminal side or has only had identifiers added, differential privacy noise of appropriate intensity is injected on the cloud side according to its corresponding remaining privacy budget.

[0055] Based on the data value attributes, privacy budget consumption, and data quality confidence level of the data uploaded by each user terminal, a dynamic aggregation weight is calculated for each user's contribution data, using the following formula: ; In the formula, For the final calculated user Dynamic aggregation weights; For users The data value attributes of the uploaded data; This is the sum of the data value attributes of all users participating in this aggregation round; For users This is the privacy budget consumed for this upload; Sensitivity coefficient for privacy consumption; For users Confidence level of the quality of the uploaded data.

[0056] Using dynamic aggregation weights, a weighted average is calculated on all privacy-post-processed user data or embedding vectors to generate the current round's aggregation update.

[0057] The aggregated update values ​​are then incorporated into the parameters of the current global federated recommendation model through secure multi-party computation or homomorphic encryption to complete the model update iteration.

[0058] The updated global model's partial metadata or the initial parameters required for the next round of training are securely distributed to the user terminals participating in the collaboration.

[0059] Specifically, based on the updated global federated recommendation model, personalized recommendation content is generated and pushed to the corresponding user terminals.

[0060] Based on the user's feedback and interaction data regarding the pushed content, the data value attributes calculated for the subsequent time for that device are dynamically adjusted, and the adjusted data value attributes are used in the local joint state information of subsequent rounds.

[0061] Preferably, by designing a cloud aggregation protocol that can be compatible with and securely aggregate heterogeneous data forms (such as original sequences and noisy embeddings), and introducing a dynamic weight calculation mechanism based on data value, privacy consumption, and quality confidence, the core challenges of heterogeneous component fusion and contribution quality measurement in open federated learning are solved. An incentive-compatible contribution evaluation system is established, which effectively guides participant behavior by assigning higher aggregation weights to high-value and privacy-protected data, while enhancing the model's robustness to low-quality or malicious contributions. In addition, this mechanism realizes knowledge fusion across different data representation modalities, laying the algorithmic foundation for building a more powerful and general federated model.

[0062] For example, the cloud server simultaneously receives different data packets from multiple terminals, such as a noisy embedding vector uploaded by terminal A and a homomorphically encrypted raw data fragment uploaded by terminal B. The cloud first decodes and verifies each data packet, and injects noise corresponding to the remaining privacy budget into the data of terminal B. Then, based on the value attributes, privacy consumption, and data quality of each data packet, it calculates dynamic aggregation weights and performs a weighted average to generate the model update gradient. Finally, the gradient is updated to the global federated recommendation model through a secure aggregation algorithm, and the updated model parameter summary is sent to the participating terminals.

[0063] In summary, this invention systematically solves the challenge of balancing privacy protection, resource consumption, and model quality in federated recommendations by introducing an integrated collaborative decision-making mechanism based on real-time terminal status, dynamically adapting to data processing methods and privacy budget allocation. It allows heterogeneous terminals to participate and contribute value in a differentiated manner, constructing a resilient privacy protection chain for end-to-cloud collaboration, significantly improving the system's robustness and overall performance in complex environments. Through closed-loop optimization of core parameters driven by user feedback, it achieves system self-evolution and continuous improvement in recommendation accuracy, thus providing a secure, efficient, and adaptive integrated solution for personalized services in big data environments.

[0064] Example 2, an embodiment of the present invention, provides an intelligent recommendation system for big data, comprising: an information acquisition module for acquiring local federated state information of user terminals; an instruction generation module for generating an integrated collaborative instruction based on the local federated state information, wherein the instruction synchronously determines the data processing form and privacy budget allocation scheme; a data upload module for user terminals to process local user data in a corresponding form and add noise according to the integrated collaborative instruction, and upload the results; and a model update module for a cloud server to receive differentiated data from various user terminals, perform secure aggregation, and update the global federated recommendation model.

[0065] Example 3 is an embodiment of the present invention, which differs from the previous embodiment in that: If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0066] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0067] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0068] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0069] 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 it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for intelligent recommendation and personalized push based on big data, characterized in that: include: Obtain the local union status information of the user terminal; Based on the local joint state information, an integrated collaborative instruction is generated, which simultaneously determines the data processing format and privacy budget allocation scheme. The user terminal processes local user data and adds noise according to the integrated collaborative instructions, and then uploads the results. The cloud server receives differentiated data from various user terminals, performs secure aggregation, and updates the global federated recommendation model.

2. The intelligent recommendation and personalized push method for big data as described in claim 1, characterized in that: The information includes computing resource status, network status, data value assessment of user data to be uploaded, and user privacy preferences; The data processing method specifies uploading one of the following to the cloud: raw data, locally generated embedded vectors, or locally encrypted data. The privacy budget allocation scheme dynamically allocates the budget ratio for adding differential privacy noise on the terminal side and the cloud side within a preset total privacy budget.

3. The intelligent recommendation and personalized push method for big data as described in claim 2, characterized in that: The specific details of obtaining the local federated state information of the user terminal are as follows: The system collects in real time the computing resource status, network connection status, data value attributes of user behavior data to be uploaded, and the privacy preference level actively set by the user. The computing resource status includes at least the currently available computing power and the remaining power. The network connection status includes at least uplink bandwidth and transmission latency; The data value attributes are determined by analyzing the novelty, sparsity, and potential contribution of the user behavior data to the global model.

4. The intelligent recommendation and personalized push method for big data as described in claim 3, characterized in that: Based on the local joint state information, an integrated collaborative instruction is generated. This instruction simultaneously determines the data processing format and the privacy budget allocation scheme, as detailed below: The computing resource status, network connection status, data value attributes, and user privacy preference level are input into a lightweight decision model; The lightweight decision model simultaneously outputs the selection result of the data processing form and the privacy budget allocation scheme; The choice of the data processing method is based on a comprehensive trade-off between data upload timeliness, computational overhead, and privacy exposure risks; The privacy budget allocation scheme is dynamically generated based on a comprehensive assessment of the local resource consumption of adding noise on the terminal side and the communication privacy risks of adding noise on the cloud side.

5. The intelligent recommendation and personalized push method for big data as described in claim 1, characterized in that: The user terminal processes local user data and adds noise according to the integrated collaborative instructions, and then uploads the results. The specific content is as follows: Based on the data processing format specified in the integrated collaborative instruction, the local edge computing module is invoked to perform the corresponding operation: When the data is uploaded in the form of raw data, the original user behavior sequence to be uploaded is directly cached. When the form is an uploaded locally generated embedding vector, the original user behavior sequence is encoded into a short-term interest embedding vector using a local lightweight neural network model; When the form is uploading locally encrypted data, homomorphic encryption or secure multi-party computation preprocessing is performed on the original user behavior sequence or the generated embedding vector. Based on the privacy budget allocation scheme specified in the integrated collaborative instruction, differential privacy noise that meets the budget requirements is injected into the corresponding data processing stage: If the budget allocation scheme indicates that noise should be added on the terminal side, then Laplace noise or Gaussian noise of appropriate intensity should be added to the cached, encoded or encrypted data. If the budget allocation scheme indicates that noise addition is delayed in whole or in part to the cloud, then add weak noise or only add identifiers to the data to be uploaded; The data, after morphological processing and noise injection, is encapsulated into data packets according to a preset secure communication protocol and uploaded to the cloud server via a network connection.

6. The intelligent recommendation and personalized push method for big data as described in claim 1, characterized in that: The cloud server receives differentiated data from various user terminals, performs secure aggregation, and updates the global federated recommendation model, as detailed below: The system receives data packets with different data processing formats from different user terminals in parallel, and decodes and verifies the data packets. For data packets identified as raw data formats, user behavior sequences are extracted directly; For data packets identified as embedding vectors, extract noisy short-term interest embedding vectors; For data packets identified as locally encrypted data, secure decoding is performed using the corresponding key or protocol; Data with fully added differential privacy noise is used directly for aggregation; For data that has not had sufficient noise added on the terminal side or has only had identifiers added, differential privacy noise of corresponding strength is injected on the cloud side according to its corresponding remaining privacy budget. Based on the data value attributes, privacy budget consumption, and data quality confidence of the data uploaded by each user terminal, a dynamic aggregation weight is calculated for each user's contribution data. Using the dynamic aggregation weights, a weighted average is calculated on all privacy-post-processed user data or embedding vectors to generate the current round of aggregation update. The aggregated update amount is then incorporated into the parameters of the current global federated recommendation model through secure multi-party computation or homomorphic encryption to complete the model update iteration; The updated global model's partial metadata or the initial parameters required for the next round of training are securely distributed to the user terminals participating in the collaboration.

7. The intelligent recommendation and personalized push method for big data as described in claim 6, characterized in that: Based on the updated global federated recommendation model, personalized recommendation content is generated and pushed to the corresponding user terminals. Based on the user terminal's feedback and interaction data regarding the pushed content, the data value attribute calculated for the terminal in subsequent rounds is dynamically adjusted, and the adjusted data value attribute is used in the local joint state information of subsequent rounds.

8. An intelligent recommendation system for big data, based on the intelligent recommendation and personalized push method for big data as described in any one of claims 1 to 7, characterized in that: include, The information acquisition module is used to acquire the local joint status information of the user terminal; The instruction generation module is used to generate an integrated collaborative instruction based on local joint state information. This instruction simultaneously determines the data processing format and privacy budget allocation scheme. The data upload module is used by the user terminal to process local user data in a corresponding form and add noise according to the integrated collaborative instructions, and then upload the results. The model update module is used by the cloud server to receive differentiated data from various user terminals, perform secure aggregation, and update the global federated recommendation model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the intelligent recommendation and personalized push method for big data as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the intelligent recommendation and personalized push method for big data as described in any one of claims 1 to 7.

Images