Image generation method and system based on privacy calculation

Abstract

The invention provides a portrait generation method and system based on privacy calculation, and relates to the technical field of data processing, and the method comprises the steps: carrying out the data collection of a target user through calling a plurality of data channels, and obtaining a user multi-party data set; traversing the data set to carry out local learning training to obtain a local learning log, carrying out security encryption on the data set to determine an encrypted data set, transmitting the encrypted data set to a central service holder to carry out privacy calculation to generate a data aggregation result, carrying out result decryption according to the aggregation result, and updating the data set to generate an update result; and synchronizing the updating result to a plurality of data channels to carry out portrait modeling to generate comprehensive portrait information of the target user and formulate recommendation suggestions, so that the problems of low data processing efficiency and insufficient privacy protection of a portrait generation method are solved; and the effects of efficiently generating the comprehensive portrait of the target user and formulating personalized recommendation suggestions according to the comprehensive portrait on the premise of ensuring the privacy security of the user data are achieved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and specifically to a method and system for generating profiles based on privacy computing. Background Technology

[0002] In the wave of digital transformation, the in-depth mining and application of user data has become an important means for many industries to improve service quality and user experience. Analyzing user data can generate user profiles, enabling personalized recommendations to improve personalized services and customer conversion rates. However, the user data involved in this process often contains personal privacy information. How to efficiently and securely utilize this data while protecting user privacy has become a pressing technical challenge. While some privacy-focused profile generation methods exist, these methods often suffer from low data processing efficiency and insufficient privacy protection. For example, some methods fail to fully consider data sensitivity and privacy when processing data, increasing the risk of data leakage, while others suffer from low processing efficiency due to complex data processing procedures and high computational loads, failing to meet the needs of practical applications.

[0003] In summary, existing portrait generation methods often suffer from technical problems such as low data processing efficiency and insufficient privacy protection. Summary of the Invention

[0004] This application provides a privacy-based image generation method and system to address the technical problems of low data processing efficiency and insufficient privacy protection in existing image generation methods.

[0005] In view of the above problems, this application provides a method and system for generating portraits based on privacy computing.

[0006] In a first aspect, this application provides a method for generating profiles based on privacy-preserving computation, the method comprising:

[0007] Multiple data channels are accessed to collect data from the target user, resulting in a multi-party user dataset. This dataset is then traversed for local learning training to obtain local learning logs. The dataset is then securely encrypted according to these logs to determine the encrypted dataset. This encrypted dataset is transmitted to a central service cloud platform for privacy computation, generating a data aggregation result. The aggregation result is then decrypted to update the multi-party user dataset, generating an updated data result. This updated data result is synchronized to the multiple data channels for profile modeling to generate a comprehensive profile of the target user. Recommendations are then developed based on this comprehensive profile information.

[0008] Secondly, this application provides a profile generation system based on privacy computing, the system comprising:

[0009] The system comprises the following modules: a data acquisition module, which accesses multiple data channels to collect data from the target user and obtain a multi-party user dataset; a learning and training module, which iterates through the multi-party user dataset to perform local learning and training, obtaining local learning logs; a privacy computation module, which securely encrypts the multi-party user dataset according to the local learning logs, determines the encrypted dataset, transmits the encrypted dataset to the central service cloud platform for privacy computation, and generates a data aggregation result; a data update module, which decrypts the data aggregation result and updates the multi-party user dataset to generate a data update result; and a profile modeling module, which synchronizes the data update result to the multiple data channels to perform profile modeling, generate comprehensive profile information for the target user, and formulates recommendation suggestions based on the comprehensive profile information.

[0010] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0011] This application provides a privacy-preserving computation-based user profile generation method. This method involves accessing multiple data channels to collect data from target users, obtaining a multi-party user dataset. The multi-party user dataset is then traversed for local learning training to obtain local learning logs. The dataset is then securely encrypted according to these logs, resulting in an encrypted dataset. This encrypted dataset is transmitted to a central service cloud platform for privacy computation, generating a data aggregation result. The aggregation result is then decrypted to update the multi-party user dataset, generating an updated data result. This updated data result is synchronized to the multiple data channels for profile modeling, generating a comprehensive user profile. Based on this comprehensive profile, recommendations are developed. This method addresses the technical problems of low data processing efficiency and insufficient privacy protection in existing user profile generation methods, achieving the technical effect of efficiently generating a comprehensive user profile and developing personalized recommendations while ensuring user data privacy and security. Attached Figure Description

[0012] Figure 1 This application provides a schematic diagram of a portrait generation method based on privacy computing.

[0013] Figure 2 This application provides a schematic diagram of the structure of a privacy-based image generation system.

[0014] Figure labeling: Data acquisition module 11, learning and training module 12, privacy computing module 13, data update module 14, profile modeling module 15. Detailed Implementation

[0015] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0016] Example 1, as Figure 1 As shown, this application provides a method for generating profiles based on privacy computing, the method comprising:

[0017] Multiple data channels are accessed, and data is collected from the target user through these channels to obtain a multi-party dataset of the user.

[0018] Privacy-preserving computation refers to a technology that performs data analysis and computation while ensuring data privacy. It ensures data security and privacy by introducing various encryption and anonymization techniques during data processing, while also guaranteeing the accuracy and effectiveness of the data analysis results. Profile generation, on the other hand, refers to generating user profiles reflecting user characteristics, preferences, and behavioral patterns through in-depth mining and analysis of user data. These profiles can provide enterprises with valuable user insights, helping them better understand user needs and behaviors, thereby developing more precise personalized service strategies. This application proposes a privacy-preserving computation-based profile generation method that aims to improve data processing efficiency while strengthening data privacy protection, thus addressing the problems of low data processing efficiency and insufficient privacy protection in existing profile generation methods.

[0019] Specifically, multiple data channels are accessed, each connected to different data sources, including but not limited to social media platforms and user shopping behavior data recording systems. This diversity ensures the comprehensiveness and richness of the collected user data. The multiple data channels refer to the multiple connection paths established for data collection, each corresponding to a different data source and capable of transmitting data in parallel or serially, thereby improving the efficiency and flexibility of data collection. Furthermore, each data channel is equipped with a corresponding data interface and data transmission protocol to ensure accurate and efficient data transmission.

[0020] After data collection, a multi-faceted dataset of users is obtained. This dataset consists of data from various channels, including user behavioral characteristics (such as browsing, clicking, and purchasing records), demographic characteristics (such as age, gender, and occupation), purchase history (i.e., user purchase records on purchasing platforms), and browsing history (user browsing footprints on web pages or apps), among other multi-dimensional information. This information collectively constitutes the raw data for user profiles, providing a rich data foundation for subsequent user profile generation. For example, social media platform data channels can yield user social behavior, interests, and points of interest; shopping behavior data systems can provide browsing history and search records. This data is collected and aggregated through its respective channels to form a complete multi-faceted user dataset.

[0021] In summary, by accessing multiple data channels and collecting data from target users from multiple dimensions and perspectives, a comprehensive and rich multi-faceted user dataset was obtained, providing solid data support for subsequent user profile generation and personalized recommendation formulation.

[0022] The local learning training is performed by traversing the user multi-party dataset to obtain local learning logs.

[0023] Furthermore, the potential information and features within the user dataset are deeply mined to provide more refined data support for subsequent user profiling. Each data record in the user multi-party dataset is examined and processed individually. This step ensures that all information in the dataset is fully utilized and incorporated into subsequent learning and training without omission. During the traversal, a local learning training method is employed. This method is designed for small-scale or specific datasets, aiming to extract useful features and patterns from local data. Compared to global learning training, local learning training is more flexible, allowing for targeted analysis and processing of different parts of the data, thereby improving learning efficiency and accuracy. When performing local learning training, each subset or specific type of data in the user multi-party dataset, such as browsing history data, is independently trained. This process utilizes various machine learning algorithms and models, such as decision trees, neural networks, and cluster analysis, to extract key features and patterns from the data. For example, association rule learning algorithms can be used for purchase history data to discover potential associations and patterns in user purchasing behavior; time series analysis models can be used for browsing history data to capture the temporal patterns and trends of user browsing behavior. The results of these local learning training sessions are recorded in local learning logs, serving as an important reference for subsequent user profile construction. The local learning logs detail the process of each local learning training session, the algorithms and models used, and the key features and patterns extracted. These logs not only provide data support for subsequent user profile construction but also facilitate the evaluation and optimization of the learning training effectiveness.

[0024] In summary, by traversing multiple user datasets and performing local learning training, we successfully obtained local learning logs containing rich features and patterns, laying a solid foundation for subsequent user profile construction and personalized recommendation formulation.

[0025] The user's multi-party dataset is securely encrypted according to the local learning logs, the encrypted dataset is determined, and the encrypted dataset is transmitted to the central service cloud platform for privacy computing to generate data aggregation results.

[0026] Optionally, secure encryption refers to using encryption technologies and algorithms to encrypt sensitive information in user multi-party datasets to prevent unauthorized access or leakage during transmission and storage. This step is crucial for ensuring user privacy and data security. Appropriate encryption algorithms and key management strategies are selected based on the data characteristics and importance recorded in the local learning logs to ensure encryption strength and effectiveness. During encryption, special attention must be paid to data integrity and availability. By employing advanced encryption technologies, such as homomorphic encryption and differential privacy, necessary data processing and analysis can be performed without exposing the original data. These encryption technologies ensure that data can still be effectively utilized in an encrypted state, thus meeting the needs of subsequent privacy-preserving computations.

[0027] After encryption, encrypted datasets are obtained, which retain the original data value while possessing enhanced privacy protection capabilities. These encrypted datasets are then securely transmitted to the central service cloud platform. As the core node for data processing, the central service cloud platform possesses powerful computing and data storage capabilities, enabling efficient processing and analysis of large-scale datasets. Advanced privacy-preserving computation techniques are employed on the central service cloud platform to perform in-depth processing of the encrypted datasets. Privacy-preserving computation can extract useful information and features from the data without exposing the original data. Appropriate privacy-preserving computation algorithms and models, such as secure multi-party computation and federated learning, are selected based on actual needs to process and analyze the encrypted datasets. Finally, data aggregation results are generated through privacy-preserving computation techniques. These data aggregation results are the result of comprehensive analysis and mining of user data from multiple dimensions and perspectives while protecting user privacy. They not only reflect user behavioral characteristics and preferences but also reveal potential correlations and patterns between data, providing strong support for subsequent user profile construction and personalized recommendation formulation.

[0028] In summary, secure encryption processing, encrypted dataset transmission, and privacy-preserving computation on the central service cloud platform successfully generated data aggregation results, providing a reliable data foundation for subsequent user profile construction and personalized recommendation formulation.

[0029] Based on the data aggregation results, the results are decrypted and the user multi-party dataset is updated to generate data update results.

[0030] For example, the data aggregation described above involves merging data from different data sources according to specific rules or standards to form a unified dataset. This aggregation may be based on various factors such as user ID, timestamps, and geographic location. The aggregated dataset can more comprehensively reflect user behavior and characteristics, providing strong support for subsequent analysis. After data aggregation, the results often contain encrypted or encoded information, a measure taken to protect user privacy and data security. Therefore, result decryption becomes a necessary step. The decryption process requires the use of specific decryption algorithms or keys to restore the encrypted information in the aggregation results to the original data or readable information. This step is crucial for subsequent data updates and analysis.

[0031] Next, the user multi-party dataset is updated. The purpose of this step is to compare and integrate the decrypted data aggregation results with the original user multi-party dataset. Updates may include newly added data records, modified data items, or deleted outdated data. The update process must ensure data accuracy and consistency, avoiding duplication or omissions. For example, suppose the original user multi-party dataset contains information such as user age, gender, and purchase history. After data aggregation, it may contain the user's latest purchase history and browsing behavior. In this case, it is necessary to compare this new information with the original dataset and update the user's purchase history and behavioral characteristics. If the user's age or gender is updated or confirmed in the aggregation results, the relevant information in the original dataset also needs to be modified accordingly. The updated dataset will contain more complete and accurate information, providing a more reliable foundation for subsequent data analysis and decision-making. The output of this step may be a new dataset file, a database update record, or a report containing updated information.

[0032] In summary, the process of decrypting the data aggregation results, updating the user multi-party dataset, and generating updated data results is a complex and sophisticated data processing workflow. It involves multiple stages such as data collection, aggregation, decryption, updating, and output, and each stage requires strict control and verification to ensure the accuracy and security of the data.

[0033] The data update results are synchronized to the multiple data channels to generate comprehensive profile information of the target user through profile modeling, and recommendation suggestions are made based on the comprehensive profile information.

[0034] Specifically, the aggregated, decrypted, and updated dataset is accurately transmitted to multiple pre-defined data channels. These channels may include databases, data warehouses, and real-time data streaming platforms, which together form a data flow network. During synchronization, it is crucial to ensure data integrity, consistency, and timeliness, avoiding data loss, distortion, or delays during transmission. For example, suppose there is a data update containing users' latest purchase records, browsing behavior, and preference information. To synchronize this information to the data warehouse, ETL (Extract, Transform, Load) tools may be used to extract the data from the source system, perform necessary cleaning and transformation, and finally load it into the corresponding tables in the data warehouse.

[0035] Next, user profile modeling will be performed. In this step, data synchronized to various data channels will be combined with advanced technologies such as machine learning and data mining to construct user profile models. These models can comprehensively and deeply depict a user's overall profile based on multiple attributes (such as age, gender, occupation, and interests) and behaviors (such as purchase history, browsing history, and interaction patterns). Taking user purchase behavior as an example, key characteristics such as purchase preferences, spending power, and purchase frequency can be identified by analyzing a user's purchase records within a specific time period. Simultaneously, combining data such as browsing history and click behavior can further uncover the user's potential needs and interests. Generating comprehensive user profile information is a direct result of user profile modeling. This profile information may be presented in various forms, such as user tags, user profile reports, and user profile cards. These not only contain basic user attribute information but also incorporate personalized characteristics and preference information, providing rich data support for subsequent applications of user profiles.

[0036] Finally, recommendations are made based on the comprehensive user profile information. After obtaining the user's comprehensive profile, recommendation algorithms, collaborative filtering, and other technologies can be used to tailor personalized recommendations for the user. These recommendations may cover multiple aspects such as services and content, aiming to meet the user's individual needs and interests. For example, for a user who likes to buy fashion clothing and is sensitive to trends, the latest fashion items, styling suggestions, and trend analysis can be recommended based on their comprehensive profile information. Such recommendations not only help improve the user's shopping experience but also enhance user stickiness and loyalty.

[0037] In summary, the process of synchronizing data update results to multiple data channels for profile modeling and formulating recommendations based on comprehensive profile information involves multiple stages, including data synchronization, modeling, analysis, and application. Each stage requires strict control and optimization to ensure the accuracy and effectiveness of the final recommendations.

[0038] Furthermore, the process of traversing the user multi-party dataset for local learning training to obtain local learning logs includes: setting a desired training objective; performing data correlation analysis on the user multi-party dataset according to the desired training objective to generate multiple correlation coefficients; performing local filtering on the user multi-party dataset according to the multiple correlation coefficients to determine a filtered data group; performing association training on the filtered data group to construct a judgment learner; and synchronizing the user multi-party dataset to the judgment learner for judgment to generate the local learning logs.

[0039] Furthermore, a clear expected training objective is first set, which serves as the overall direction and foundation of the entire learning and training process. For example, if the expected training objective is to generate user profiles for purchasing, then the focus will be on considering user purchasing preferences or data related to purchasing behavior. This expected training objective provides clear guidance for subsequent data processing and analysis. Next, data correlation analysis is performed on the user's multi-faceted dataset according to the set expected training objective. This step aims to identify which data items are correlated and the strength of these correlations. By using statistical methods or machine learning algorithms, multiple correlation coefficients can be calculated, which quantify the strength of the correlation between data items. For example, a high correlation might be found between a user's browsing history and purchase history, while the correlation between a user's age and purchasing preferences might be relatively low.

[0040] After obtaining multiple correlation coefficients, the user multi-dataset is then locally filtered according to these coefficients. The purpose of this step is to select data items from the original dataset that are most relevant to the desired training objective and most valuable for building the judgment learner, thus forming a filtered data group. The filtering can be based on the magnitude of the correlation coefficients or other rules set according to the specific business scenario. For example, several data items with the highest correlation to purchasing preferences, such as browsing history, purchase history, and search history, can be selected to form the filtered data group. Then, the filtered data group is subjected to association training. This step uses machine learning algorithms to train the selected data items to build a judgment learner capable of accurately judging user profiles. During training, the algorithm learns the relationships between data items and attempts to find the feature combinations that best describe the user profile. For example, the algorithm might learn that a user's browsing history and purchase history jointly determine their purchasing preferences, thus using these features as important components of the judgment learner.

[0041] Finally, the user multi-party dataset is synchronized to the judgment learner for evaluation. This step is crucial for verifying the performance of the judgment learner. By inputting the original dataset into the judgment learner, the learner's prediction results for user profiles can be observed. Furthermore, based on the accuracy and stability of the prediction results, a local learning log can be generated, recording the training process and performance of the judgment learner. The local learning log not only contains details of the judgment learner's construction but also reflects changes in the dataset during training and the learner's optimization process, providing important reference for subsequent user profile generation.

[0042] Furthermore, the process of securely encrypting the user multi-party dataset according to the local learning logs to determine the encrypted dataset includes: performing data sensitivity analysis on the user multi-party dataset based on the local learning logs to determine multiple sensitivities; traversing and identifying the user multi-party dataset according to the multiple sensitivities to generate multiple sensitive datasets; collecting data environment information of the user multi-party dataset, continuously monitoring the data environment information, and generating an environmental change monitoring parameter set; determining whether the environmental change monitoring parameter set is within the environmental stability range; if the environmental change monitoring parameter set is within the environmental stability range, generating an environmental security identifier, encrypting the multiple sensitive datasets according to the environmental security identifier, and generating an encrypted dataset; if the environmental change monitoring parameter set is not within the environmental stability range, generating an environmental anomaly identifier, issuing an intrusion alarm according to the environmental anomaly identifier, and generating an alarm prompt; feeding back the alarm prompt to the environmental change monitoring parameter set to encrypt and filter the multiple sensitive datasets, encrypting the filtered data, and generating the encrypted dataset.

[0043] Optionally, data sensitivity analysis is performed on the user multi-party dataset based on local learning logs. This step aims to identify the sensitivity level of each data item in the dataset. Sensitivity levels are categorized based on the potential risk of data breach. For example, user identity information and payment information are typically considered highly sensitive data, while user browsing history and preference information may be classified as moderately or lowly sensitive. By deeply analyzing the data usage patterns, access frequency, and correlations with other data recorded in the local learning logs, the sensitivity of each data item can be accurately determined. Next, the user multi-party dataset is traversed according to the determined sensitivity levels, and each data item is labeled with a corresponding sensitivity tag, thereby generating multiple sensitive datasets. These sensitive datasets are managed separately according to their sensitivity levels to facilitate the implementation of different levels of security measures subsequently.

[0044] To ensure the security of the data encryption process, further information about the data environment of multiple user datasets is collected, including but not limited to network, physical, and data storage environments. Continuous monitoring of this data environment information allows for the real-time acquisition of environmental change monitoring parameters, reflecting the current state and stability of the data environment. Subsequently, the system determines whether the environmental change monitoring parameter set is within a safe and stable range based on a preset environmental stability interval standard. This environmental stability interval standard can be adaptively set. If the parameter set is within the environmental stability interval, it indicates that the current data environment is relatively secure, and data encryption can be performed safely. At this point, the system generates an environmental security identifier and uses an appropriate encryption algorithm to encrypt multiple sensitive datasets based on this identifier, ultimately generating an encrypted dataset. The encryption algorithm can be symmetric, asymmetric, or homomorphic, such as AES symmetric encryption, which is suitable for efficiently encrypting large amounts of sensitive data; RSA asymmetric encryption, used to ensure the security of critical data transmission; and homomorphic encryption, which can be considered under specific requirements to support direct computation and processing in the encrypted state.

[0045] However, if the environmental change monitoring parameter set fails to fall within the stable environmental range, it indicates a potential security threat or anomaly in the data environment. In this case, an environmental anomaly identifier is immediately generated, triggering an intrusion alarm mechanism and generating an alert. This alert is sent to the system administrator or security team in real time, allowing them to take timely countermeasures. Simultaneously, to mitigate risks during encryption, the system filters data from multiple user datasets that may be affected by the environment based on the alert. Specifically, encryption is delayed for data items significantly impacted by environmental anomalies, while the remaining data is encrypted according to the normal process. This encryption filtering strategy ensures maximum data security even in unstable environments, while avoiding performance degradation from unnecessary encryption operations.

[0046] In summary, by combining local learning logs with real-time monitoring of data environment information, efficient and secure encryption processing of user multi-party datasets was achieved, effectively improving the level of data protection.

[0047] Furthermore, transmitting the encrypted dataset to the central service cloud platform for privacy computation to generate data aggregation results includes: compressing the encrypted dataset to generate multiple compressed data packages; constructing a secure channel and transmitting the multiple compressed data packages to the central service cloud platform through the secure channel, generating a reception confirmation signal; upon receiving the reception confirmation signal, activating the central service cloud platform to perform multi-party secure computation to generate data privacy computation results; and performing regression analysis on the data privacy computation results to generate the data aggregation results.

[0048] For example, to optimize transmission efficiency and reduce bandwidth consumption, the encrypted dataset is compressed. This step utilizes advanced compression algorithms, such as Huffman coding or the Lempel-Ziv-Welch (LZW) algorithm, to convert the original encrypted dataset into multiple compressed data packets. These compressed packets not only significantly reduce the data size but also preserve the integrity and encryption state of the original data, ensuring data security during transmission.

[0049] Next, to ensure the privacy and integrity of the data during transmission, a secure channel is established. This secure channel, based on encryption protocols (such as TLS / SSL) and authentication mechanisms, ensures that the data will not be intercepted or tampered with by unauthorized third parties during transmission. Through this secure channel, multiple data compressed packets are securely transmitted to the central service platform. Once the data compressed packets arrive successfully, the central service platform immediately generates a reception confirmation signal as feedback that the data transmission was successful.

[0050] Upon receiving the confirmation signal, the central service platform immediately activates its multi-party secure computation module. This module utilizes advanced cryptographic techniques, such as homomorphic encryption or secret sharing protocols, to allow computation and processing of data without decryption. In this way, the central service platform can perform efficient data privacy computations on multiple compressed data packages while protecting data privacy, generating data privacy computation results. These results reflect the correlation and statistical characteristics between encrypted datasets, but do not disclose any specific personal information or sensitive data.

[0051] Finally, to extract valuable information from the data privacy computation results and generate data aggregation results, regression analysis is employed. Regression analysis is a statistical method used to study the relationship between one or more independent variables and a dependent variable. In this embodiment, regression analysis is applied to the data privacy computation results to identify trends, patterns, and correlations within the dataset. Through this process, the system can generate data aggregation results that not only reveal the overall characteristics of the dataset but also provide strong support for subsequent data analysis and decision-making.

[0052] In summary, through a series of steps including data compression, secure channel transmission, multi-party secure computation, and regression analysis, efficient and secure transmission and privacy-preserving computation of encrypted datasets are achieved, providing strong technical support for data aggregation.

[0053] Furthermore, upon receiving the confirmation signal, activating the central service cloud platform to perform multi-party secure computation and generate a data privacy computation result includes: generating an activation command when the encrypted channel receives the confirmation signal; activating the computation module of the central service cloud platform through the activation command; analyzing the multiple data compressed packages through the computation module to determine the data distribution pattern; performing multi-party loading on the multiple data compressed packages according to the data distribution pattern to obtain multiple encrypted data fragments; performing key matching verification on the multiple encrypted data fragments to generate a data encryption verification result; dispatching tasks according to the data encryption verification result and a multi-party secure computation protocol to determine multiple computation sub-tasks; performing secure computation through the multiple computation sub-tasks to generate multiple intermediate results; and summarizing and integrating the multiple intermediate results to obtain the data privacy computation result.

[0054] In one specific embodiment, when the encrypted channel successfully receives a reception confirmation signal from the central service cloud platform, the system immediately generates an activation command. This activation command is the key signal that triggers the startup of the central service cloud platform's computing module, ensuring that subsequent computational processing only begins after the data has arrived securely and been confirmed to be correct. Then, through this activation command, the central service cloud platform's computing module is successfully activated. The activated computing module first analyzes the received multiple data compressed packages to determine the data distribution pattern. The data distribution pattern refers to the relative frequency and distribution of various elements or categories in the data, which is the basis for subsequent data processing and computation. Through this step, the system can better understand the structure and characteristics of the data, preparing for subsequent multi-party loading and computation. Subsequently, the multiple data compressed packages are multi-party loaded according to the data distribution pattern. Multi-party loading refers to distributing data to multiple participating nodes, i.e., different parts of the computing module or different computing resources, for parallel processing. In this process, each data compressed package is split into multiple encrypted data fragments and distributed to different participating nodes. These encrypted data fragments remain encrypted during transmission and storage, ensuring data privacy.

[0055] After loading multiple encrypted data fragments, the next step is key matching verification. Key matching verification involves verifying the key of each encrypted data fragment to ensure it matches the expected key, thereby ensuring the integrity and authenticity of the data. The expected key is pre-agreed or generated to ensure the confidentiality, integrity, and authenticity of the data. Before encrypted communication begins, the communicating parties typically agree on one or more keys, which can be symmetric keys (such as AES keys) or asymmetric key pairs (such as RSA public and private keys). The expected key is these pre-agreed keys, which play a crucial role in the subsequent encryption and decryption processes. During encrypted communication, key matching verification is usually achieved by comparing the hash values ​​of the keys, using key exchange protocols (such as the Diffie-Hellman protocol), or digital signatures. This step is a critical part of the data privacy computation process, effectively preventing data from being tampered with or replaced during transmission or storage.

[0056] Following the key matching verification, the system distributes tasks according to a multi-party secure computation protocol. This protocol allows multiple participants to collaboratively compute function results without revealing their private data. In this step, the system breaks down the computation task into multiple sub-tasks based on the data distribution pattern and the number of encrypted data fragments, assigning them to different participating nodes. Each node only operates on its assigned encrypted data fragment and cannot access the complete data, thus ensuring data privacy. Then, through the execution of multiple sub-tasks, the system begins secure computation. Secure computation refers to the process of computation and processing while protecting data privacy. In this step, each participating node performs computations on its assigned encrypted data fragments according to its assigned sub-task, generating intermediate results. These intermediate results remain encrypted during transmission and storage to ensure data privacy. Finally, the multiple intermediate results are aggregated and integrated. Aggregation and integration refers to merging and processing the intermediate results generated by each participating node to generate the final data privacy computation result. This step is the final step in the data privacy computation process; by merging and processing the intermediate results, it yields a data privacy computation result that reflects the overall characteristics and trends of the data.

[0057] Furthermore, synchronizing the data update results to the multiple data channels for profile modeling to generate comprehensive profile information for the target user includes: extracting features based on the data update results to determine multiple update features; identifying and dividing the multiple data channels according to the multiple update features to determine multiple data matching tags; dividing the data update results into data blocks according to the multiple data matching tags to generate multiple data blocks; synchronously matching the multiple data blocks to the multiple data channels according to the multiple update features for collaborative filtering to generate data similarity coefficients; performing cluster analysis on the multiple data blocks according to the data similarity coefficients to construct clustered profiles; performing data dimensionality reduction based on the clustered profiles to obtain implicit profile features; compensating the clustered profiles based on the implicit profile features to construct dynamic profile information; and adding the dynamic profile information to the comprehensive profile information of the target user.

[0058] Furthermore, feature extraction is performed based on the obtained data update results. Feature extraction refers to extracting key information that represents the characteristics of the data from the original data; this key information is called update features. Feature extraction is achieved through advanced algorithms and techniques, such as Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA), to ensure that the extracted update features accurately reflect the latest changes and trends in the data. Next, multiple data channels are labeled and divided according to the extracted update features. The purpose of this step is to associate data channels with update features for subsequent data processing and profiling. Through this step, one or more data matching labels are assigned to each data channel, and these labels represent the type and number of update features contained in that channel. Then, the data update results are divided into data blocks according to the assigned data matching labels. Data block division refers to dividing the original data into multiple smaller data blocks according to a certain rule or algorithm for more efficient processing and analysis. In this embodiment, data block division is based on the similarity and relevance of update features to ensure that each data block contains update data with similar or related features. Next, the multiple data blocks are synchronously matched to multiple data channels according to the update features for collaborative filtering. Collaborative filtering is a recommendation algorithm that predicts content a user might be interested in by analyzing user behavior and interests. In this embodiment, collaborative filtering is used to identify the similarity and correlation between different data blocks and generate data similarity coefficients. These similarity coefficients reflect the degree of similarity between data blocks and are an important basis for subsequent clustering analysis and user profile modeling. Subsequently, clustering analysis is performed on multiple data blocks according to the generated data similarity coefficients. Clustering analysis is a data mining technique that groups data objects into multiple classes or clusters, such that data objects within the same cluster have high similarity, while data objects between different clusters have low similarity. Clustering analysis is used to construct segmented user profiles, that is, grouping data blocks with similar characteristics into the same class to form profiles of different user groups. Then, data dimensionality reduction is performed based on the constructed segmented user profiles. Data dimensionality reduction refers to transforming the high-dimensional feature space of the original data into a low-dimensional feature space through some method to reduce the complexity and redundancy of the data. Here, data dimensionality reduction is achieved through techniques such as Principal Component Analysis (PCA) or Singular Value Decomposition (SVD) to extract latent features of the user profile. These latent features are key information hidden in the original data and are crucial for constructing accurate and comprehensive user profiles. Next, the segmented profiles are compensated based on the extracted latent features. Compensation refers to revising and improving the original profiles to more accurately and comprehensively reflect the actual situation of users. Compensation is achieved by introducing new data and information, such as user behavior data and social media data, to enrich and improve the content of the segmented profiles. Finally, the compensated dynamic profile information is added to the comprehensive profile information of the target users.Comprehensive user profile information refers to a user profile generated by integrating data from multiple sources and types. It includes various user characteristics and attributes, such as interests, consumption habits, and social relationships. The addition of dynamic user profile information makes the comprehensive user profile information more complete and accurate, providing strong support for subsequent personalized recommendations and decision support.

[0059] In summary, through a series of steps including feature extraction, data segmentation, collaborative filtering, cluster analysis, data dimensionality reduction, and compensation, the latest data updates were synchronized to multiple data channels for profile modeling, generating comprehensive profile information for the target user. This process not only improved the efficiency and accuracy of data processing but also provided strong data support for subsequent personalized recommendations and decision support.

[0060] Furthermore, the recommendation suggestions based on the comprehensive profile information include: sorting user preferences according to the dynamic profile to generate a preference sequence; combining the dynamic profile with user historical behavior information according to the preference sequence to generate content recommendation data; evaluating the content recommendation data based on the user multi-party dataset to generate multiple recommendation scores; performing reinforcement learning according to the multiple recommendation scores to generate recommendation learning results; updating the content recommendation data according to the recommendation learning results to generate the recommendation suggestions.

[0061] Specifically, user preferences are sorted based on dynamic profile information to generate a preference sequence. This preference sequence is a list arranged from highest to lowest user preference level, reflecting the user's preferences and interests in different types or content. This step is achieved by analyzing key information such as user characteristics, interests, and behavioral patterns in the dynamic profile, ensuring that the preference sequence accurately reflects the user's true preferences. Next, the dynamic profile is collaboratively processed with the user's historical behavior information according to the generated preference sequence. Collaborative processing refers to fusing and matching data from different sources or types to generate more comprehensive and accurate content recommendation data. In this embodiment, collaborative processing is achieved by matching and fusing user preferences in the dynamic profile with behavioral trajectories, purchase records, browsing records, etc., in the user's historical behavior information, thereby generating content recommendation data that includes both the user's current preferences and reflects their historical behavior.

[0062] Then, the generated content recommendation data was evaluated based on a multi-user dataset. Recommendation evaluation refers to the process of verifying and assessing the accuracy and applicability of the recommended content. This evaluation was achieved by introducing a multi-user dataset, which includes user social information, geographic location information, and spending power information. By comparing and analyzing these datasets with the content recommendation data, multiple recommendation scores were generated, reflecting the accuracy and applicability of the recommended content.

[0063] Subsequently, reinforcement learning was applied based on the generated recommendation scores. Reinforcement learning is a machine learning technique that learns to make optimal decisions through continuous trial and error and optimization. It was used to optimize and adjust content recommendation data based on the recommendation scores to generate recommendation learning results. This step was achieved by building a reinforcement learning model that automatically adjusts the recommendation strategy according to changes in the recommendation scores, thereby improving the accuracy and applicability of the recommended content. Finally, the content recommendation data was updated based on the generated recommendation learning results to generate the final recommendation suggestions. These recommendations are personalized content derived from user dynamic profiles, historical behavioral information, and multi-dataset analysis, reflecting the user's true needs and preferences. By continuously updating and optimizing these recommendations, more accurate and personalized recommendation services can be provided to users, improving user satisfaction and loyalty.

[0064] In summary, by combining dynamic user profiles, historical behavioral information, and multi-source datasets, and employing a series of complex steps including preference ranking, collaborative processing, recommendation evaluation, reinforcement learning, and recommendation suggestion generation, personalized recommendation services are provided to users. This process not only improves the accuracy and applicability of recommendations but also delivers a higher quality and more personalized user experience.

[0065] Through the technical solutions of the above embodiments, the privacy-based profile generation method provided by this application solves the technical problems of low data processing efficiency and insufficient privacy protection in existing profile generation methods, and achieves the technical effect of efficiently generating a comprehensive profile of the target user and formulating personalized recommendation suggestions based on it while ensuring the privacy and security of user data.

[0066] Example 2, based on the same inventive concept as the privacy-based computation-based portrait generation method in the foregoing examples, such as... Figure 2 As shown, this application provides a privacy-preserving computation-based portrait generation system, the system comprising:

[0067] The data acquisition module 11 is used to access multiple data channels and collect data from the target user through the multiple data channels to obtain a multi-party dataset of the user.

[0068] The learning and training module 12 is used to traverse the user multi-party dataset to perform local learning and training, and obtain local learning logs.

[0069] The privacy computing module 13 is used to securely encrypt the user's multi-party dataset according to the local learning log, determine the encrypted dataset, transmit the encrypted dataset to the central service cloud platform for privacy computing, and generate data aggregation results.

[0070] The data update module 14 is used to update the user multi-party dataset by decrypting the data aggregation result and generating a data update result.

[0071] The profile modeling module 15 is used to synchronize the data update results to the multiple data channels to perform profile modeling and generate comprehensive profile information of the target user, and to formulate recommendation suggestions based on the comprehensive profile information.

[0072] Furthermore, the learning and training module 12 is also used to perform the following steps: setting a desired training objective; performing data correlation analysis on the user multi-party dataset according to the desired training objective to generate multiple correlation coefficients; performing local filtering on the user multi-party dataset according to the multiple correlation coefficients to determine the filtered data group; performing association training on the filtered data group to construct a judgment learner; synchronizing the user multi-party dataset to the judgment learner for judgment, and generating the local learning log.

[0073] Furthermore, the privacy computing module 13 is also used to perform the following steps: perform data sensitivity analysis on the user multi-party dataset based on the local learning log to determine multiple sensitivities; traverse and identify the user multi-party dataset according to the multiple sensitivities to generate multiple sensitive datasets; collect data environment information of the user multi-party dataset, continuously monitor the data environment information, and generate an environmental change monitoring parameter set; determine whether the environmental change monitoring parameter set is in a stable environmental range; if the environmental change monitoring parameter set is in the stable environmental range, generate an environmental security identifier, encrypt the multiple sensitive datasets according to the environmental security identifier, and generate an encrypted dataset; if the environmental change monitoring parameter set is not in the stable environmental range, generate an environmental anomaly identifier, perform an intrusion alarm according to the environmental anomaly identifier, and generate an alarm prompt; feed back the alarm prompt to the environmental change monitoring parameter set to encrypt and filter the multiple sensitive datasets, encrypt the filtered data, and generate the encrypted dataset.

[0074] Furthermore, the privacy computing module 13 is also used to perform the following steps: compressing the encrypted dataset to generate multiple compressed data packages; constructing a secure channel and transmitting the multiple compressed data packages to the central service cloud platform through the secure channel, generating a reception confirmation signal; upon receiving the reception confirmation signal, activating the central service cloud platform to perform multi-party secure computing and generate data privacy computing results; and performing regression analysis on the data privacy computing results to generate the data aggregation results.

[0075] Furthermore, the privacy computing module 13 is also used to perform the following steps: when the encrypted channel receives the reception confirmation signal, an activation command is generated; the activation command is used to activate the computing module of the central service cloud platform, and the computing module analyzes the multiple data compressed packages to determine the data distribution pattern; the multiple data compressed packages are loaded according to the data distribution pattern to obtain multiple encrypted data fragments; the multiple encrypted data fragments are verified by key matching to generate a data encryption verification result; according to the data encryption verification result, a task is dispatched according to a multi-party secure computing protocol to determine multiple computing sub-tasks; secure computing is performed through the multiple computing sub-tasks to generate multiple intermediate results; the multiple intermediate results are summarized and integrated to obtain the data privacy computing result.

[0076] Furthermore, the profile modeling module 15 is also used to perform the following steps: extracting features based on the data update results to determine multiple update features; identifying and dividing the multiple data channels according to the multiple update features to determine multiple data matching labels; dividing the data update results into data blocks according to the multiple data matching labels to generate multiple data blocks; synchronously matching the multiple data blocks to multiple data channels according to the multiple update features for collaborative filtering to generate data similarity coefficients; performing cluster analysis on the multiple data blocks according to the data similarity coefficients to construct clustered profiles; performing data dimensionality reduction based on the clustered profiles to obtain implicit profile features; compensating the clustered profiles based on the implicit profile features to construct dynamic profile information; and adding the dynamic profile information to the comprehensive profile information of the target user.

[0077] Furthermore, the profile modeling module 15 is also used to perform the following steps: sorting user preferences according to the dynamic profile to generate a preference sequence; combining the dynamic profile with user historical behavior information according to the preference sequence to generate content recommendation data; evaluating the content recommendation data based on the user multi-party dataset to generate multiple recommendation scores; performing reinforcement learning according to the multiple recommendation scores to generate recommendation learning results; updating the content recommendation data according to the recommendation learning results to generate the recommendation suggestions.

[0078] Through the foregoing detailed description of a privacy-based computation-based portrait generation method, those skilled in the art can clearly understand the privacy-based computation-based portrait generation system in this embodiment. As the system disclosed in the embodiment corresponds to the method disclosed in the embodiment, the description is relatively simple, and relevant parts can be referred to in the method section.

[0079] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for generating profiles based on privacy-preserving computation, characterized in that, The method includes: Multiple data channels are accessed, and data is collected from the target user through these multiple data channels to obtain a multi-party dataset of the user. The local learning training is performed by traversing the user multi-party dataset to obtain local learning logs; The user's multi-party dataset is securely encrypted according to the local learning log, the encrypted dataset is determined, and the encrypted dataset is transmitted to the central service cloud platform for privacy calculation to generate data aggregation results. Based on the data aggregation results, the results are decrypted and the user multi-party dataset is updated to generate data update results; The data update results are synchronized to the multiple data channels to generate comprehensive profile information of the target user through profile modeling, and recommendation suggestions are made based on the comprehensive profile information.

2. The portrait generation method based on privacy computing as described in claim 1, characterized in that, Local learning training is performed by traversing the user multi-party dataset to obtain local learning logs, including: Set a desired training objective, and perform data correlation analysis on the user multi-party dataset according to the desired training objective to generate multiple correlation coefficients; The user multi-party dataset is partially filtered according to the multiple correlation coefficients to determine the filtered data group; The selected data sets are subjected to association training to construct a judgment learner; The user multi-party dataset is synchronized to the judgment learner for judgment, and the local learning log is generated.

3. The portrait generation method based on privacy computing as described in claim 1, characterized in that, The user multi-party dataset is securely encrypted according to the local learning logs to determine the encrypted dataset, including: Based on the local learning logs, perform data sensitivity analysis on the user multi-party dataset to determine multiple sensitivities; The user multi-party dataset is traversed and identified according to the multiple sensitivities to generate multiple sensitive datasets; Collect data environment information from the user's multi-party dataset, continuously monitor the data environment information, and generate a set of environmental change monitoring parameters; Determine whether the set of environmental change monitoring parameters is within the environmental stability range; If the environmental change monitoring parameter set is within the environmental stability range, an environmental safety identifier is generated, and the multiple sensitive datasets are encrypted based on the environmental safety identifier to generate an encrypted dataset. If the environmental change monitoring parameter set is not within the environmental stability range, an environmental anomaly identifier is generated, and an intrusion alarm is triggered according to the environmental anomaly identifier, generating an alarm prompt. The alarm prompts are fed back to the environmental change monitoring parameter set to encrypt and filter the multiple sensitive datasets. The encrypted datasets are then encrypted based on the filtered data.

4. The portrait generation method based on privacy computing as described in claim 1, characterized in that, The encrypted dataset is transmitted to the central service cloud platform for privacy computation, generating data aggregation results, including: The encrypted dataset is compressed to generate multiple compressed data packages; A secure channel is established, and the multiple data compressed packages are transmitted to the central service cloud platform through the secure channel, generating a reception confirmation signal; Upon receiving the confirmation signal, the central service cloud platform is activated to perform multi-party security calculations and generate data privacy calculation results. The data privacy calculation results are subjected to regression analysis to generate the data aggregation results.

5. The portrait generation method based on privacy computing as described in claim 4, characterized in that, Upon receiving the confirmation signal, the central service cloud platform is activated to perform multi-party security calculations and generate data privacy calculation results, including: When the encrypted channel receives the reception confirmation signal, an activation command is generated; The activation command activates the computing module of the central service cloud platform, and the computing module analyzes the multiple data compressed packages to determine the data distribution pattern. The multiple data compressed packages are loaded from multiple sources according to the data distribution pattern to obtain multiple encrypted data fragments; The multiple encrypted data fragments are subjected to key matching verification to generate data encryption verification results; Based on the data encryption and verification results, tasks are dispatched according to a multi-party secure computation protocol to determine multiple computational sub-tasks; Secure computation is performed through the multiple computational subtasks to generate multiple intermediate results; The multiple intermediate results are aggregated and integrated to obtain the data privacy calculation result.

6. The portrait generation method based on privacy computing as described in claim 1, characterized in that, The data update results are synchronized to the multiple data channels to perform profile modeling and generate comprehensive profile information for the target user, including: Based on the data update results, feature extraction is performed to determine multiple update features; The multiple data channels are identified and divided according to the multiple update features, and multiple data matching labels are determined. The data update results are divided into data blocks according to the multiple data matching tags to generate multiple data blocks; The multiple data blocks are synchronously matched to multiple data channels according to the multiple update features for collaborative filtering, generating a data similarity coefficient; Cluster analysis is performed on the multiple data blocks according to the data similarity coefficient to construct a cluster profile; Data dimensionality reduction is performed based on the grouped profiles to obtain implicit features of the profiles. The grouped profiles are then compensated based on the implicit features of the profiles to construct dynamic profile information. The dynamic profile information is added to the target user's comprehensive profile information.

7. The portrait generation method based on privacy computing as described in claim 6, characterized in that, Recommendations are made based on the comprehensive profile information, including: Based on the dynamic profile, user preferences are sorted to generate a preference sequence; Based on the preference sequence, the dynamic profile is combined with user historical behavior information to generate content recommendation data; Based on the user multi-party dataset, the content recommendation data is evaluated to generate multiple recommendation scores. Reinforcement learning is performed based on the multiple recommendation scores to generate recommendation learning results. The content recommendation data is then updated based on the recommendation learning results to generate the recommendation suggestions.

8. A portrait generation system based on privacy computing, characterized in that, The system is used to implement the privacy-based computation-based portrait generation method according to any one of claims 1-7, the system comprising: The data acquisition module is used to access multiple data channels and collect data from the target user through these channels to obtain a multi-party dataset of the user. The learning and training module is used to traverse the user multi-party dataset to perform local learning and training, and obtain local learning logs. The privacy computing module is used to securely encrypt the user's multi-party dataset according to the local learning log, determine the encrypted dataset, transmit the encrypted dataset to the central service cloud platform for privacy computing, and generate data aggregation results. The data update module is used to update the user multi-party dataset by decrypting the data aggregation result and generating a data update result. The profile modeling module is used to synchronize the data update results to the multiple data channels to perform profile modeling and generate comprehensive profile information of the target user, and to formulate recommendation suggestions based on the comprehensive profile information.

Images