Mobile phone information privacy protection method and system based on big data analysis

Abstract

The invention discloses a mobile phone information privacy protection method and system based on big data analysis, and relates to the technical field of information security, and the method comprises the steps: collecting latest privacy protection requirements from an application market and / or a supervision department, extracting privacy protection terms, and generating a standardized privacy protection rule data set; after an encryption analysis result is obtained, a dynamic privacy protection rule file is generated in combination with big data analysis; for the dynamic privacy protection rule file, analyzing byte codes of a mobile phone application through an ASM framework, automatically modifying a sensitive API calling point, and inserting privacy monitoring logic; repackaging the modified sensitive API calling point into an APK file, and carrying out privacy monitoring verification on the application program; according to the method, encryption analysis and distributed training are adopted, safe sharing and processing of privacy data under participation of multiple parties are ensured, the protection strength of mobile phone information privacy is effectively enhanced, and the method is suitable for complex and changeable application scenes.

Description

Technical Field

[0001] The present invention relates to the field of information security technology, and in particular to a method and system for protecting mobile phone information privacy based on big data analysis. Background Art

[0002] With the popularity of smart phones, the frequency of mobile application usage has increased significantly, and the collection and use of user personal information has become increasingly complex. In order to cope with the risk of privacy leakage, how to efficiently protect user privacy data has become an important direction of current technical research. Existing privacy protection technologies mainly rely on encryption technology, access control, and permission management to prevent the leakage of user sensitive information. However, with the explosive growth of data volume and the diversification of user privacy needs, traditional privacy protection methods seem to be unable to cope with emerging privacy risks. For example, although privacy protection technology based on permission control can limit the application's access to user data to a certain extent, it cannot prevent the abuse of legitimate permissions or privacy leakage during data sharing. In addition, most of the existing encryption technologies rely on a single encryption algorithm and lack flexible adaptability to data usage scenarios. Especially in the big data environment, the demand for multi-party sharing and analysis of privacy data is increasing, and existing technologies are difficult to balance the contradiction between data availability and privacy protection.

[0003] Another shortcoming of existing technologies is that there is a lag in the formulation and implementation of privacy protection rules. The privacy protection terms and policies of mobile applications are usually defined by developers themselves, and the update cycle is long, while privacy protection regulations and standards are constantly changing, making it difficult for the implementation of privacy protection rules to keep up with the latest legal requirements. In addition, most existing privacy protection mechanisms are static rule matching, which lacks the ability to dynamically adjust in the face of increasingly complex application scenarios. For example, privacy monitoring systems usually rely on predefined rules to detect sensitive operations, but these rules are often rigid and cannot adapt to changes in user behavior and data flows in real time. Therefore, existing privacy protection technologies cannot provide sufficient flexibility and intelligent support when dealing with complex and changing privacy risks. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a mobile phone information privacy protection method based on big data analysis to solve the problems of lag and static nature of existing privacy protection rules.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a method for protecting mobile phone information privacy based on big data analysis, which includes collecting the latest privacy protection requirements from application markets and / or regulatory authorities, extracting privacy protection clauses, and generating a standardized privacy protection rule data set;

[0008] Based on the privacy-preserving rule dataset, distributed training and encrypted aggregation are used to generate global model parameters.

[0009] For global model parameters, multi-party secure computing is used to encrypt and analyze the privacy data of mobile applications;

[0010] After obtaining the encryption analysis results, generate a dynamic privacy protection rule file in combination with big data analysis;

[0011] For dynamic privacy protection rule files, the ASM framework is used to parse the bytecode of mobile applications, automatically modify sensitive API call points, and insert privacy monitoring logic;

[0012] Repackage the modified sensitive API call points into an APK file and perform privacy monitoring and verification on the application.

[0013] As a preferred solution of the mobile phone information privacy protection method based on big data analysis described in the present invention, wherein: the latest privacy protection requirements are collected from the application market and / or regulatory authorities, and privacy protection clauses are extracted to generate a standardized privacy protection rule data set. The specific steps are as follows:

[0014] Automatically obtain the latest privacy protection requirements from the application market and / or regulatory authorities through public API interfaces and web crawlers; privacy protection requirements include privacy protection terms and policies;

[0015] Based on the obtained privacy protection requirements, use NLP methods to clean up the noise data in the private text;

[0016] Use the pre-trained legal domain BERT model to perform named entity recognition on the cleaned privacy text and extract privacy protection clauses;

[0017] According to the privacy protection standards, the extracted privacy protection clauses are mapped to the clause framework to unify the format;

[0018] Based on the unified format of the clause framework, semantic similarity is used to match the regulatory clauses described in different languages;

[0019] Standardize regulatory provisions from different sources through rule mapping and hierarchical classification;

[0020] Use rule engines and knowledge graphs to verify and check the consistency of standardized regulatory clauses to form a standardized privacy protection rule data set.

[0021] As a preferred solution of the mobile phone information privacy protection method based on big data analysis described in the present invention, wherein: based on the privacy protection rule data set, distributed training and encrypted aggregation are used to generate global model parameters, and the specific steps are as follows:

[0022] Using a hierarchical federated learning architecture, the privacy-preserving rule datasets are grouped into different nodes and layers;

[0023] At each node of each layer, based on the respective data grouping, the model is trained using the local privacy rule dataset to obtain the model parameters of each layer;

[0024] Each node calculates the loss function based on the local privacy rule data set and updates the model parameters of each layer by optimizing the loss function;

[0025] The updated model parameters of each layer are sent to the central server of the layer through encrypted communication, and multi-party secure computing is used to encrypt and aggregate the parameters within the layer to output the global model parameters. The expression is:

[0026]

[0027] Among them, θ (t+1) represents the global model parameters obtained after the t+1th round of training, Indicates the weighted average of the parameters of N nodes, where N represents the total number of nodes. It represents the local model parameters of the i-th node after the t-th round of training, and t represents the number of iterations.

[0028] As a preferred solution of the mobile phone information privacy protection method based on big data analysis described in the present invention, wherein: the global model parameters are encrypted and analyzed by multi-party secure computing on the privacy data of mobile phone applications. The specific steps are as follows:

[0029] Through global model parameters, coordinate each node to perform local training without sharing original data, and update global model parameters through encrypted aggregation summary nodes;

[0030] The updated global model parameters will be fed back to each node, and each node will collect local privacy data based on the updated global model parameters;

[0031] Normalize local privacy data into a unified feature scale;

[0032] Perform homomorphic encryption on the standardized local privacy data to generate an encrypted data set;

[0033] Each node performs local model training based on its own encrypted data set and calculates encrypted gradient data;

[0034] The encrypted gradient data is transmitted to the central server, and the encrypted gradient data of each node is encrypted and aggregated through multi-party secure computing to obtain the global encrypted gradient data;

[0035] By combining the dynamic changes of the time dimension on the basis of globally encrypted gradient data, the gradient descent model parameters are updated;

[0036] Feedback the updated gradient descent model parameters to each node, and each node uses the subkey to decrypt the global gradient descent model parameters;

[0037] Use the decrypted global gradient descent model parameters and combine them with the encrypted gradient data of each node to generate encrypted analysis results.

[0038] As a preferred solution of the mobile phone information privacy protection method based on big data analysis described in the present invention, after obtaining the encryption analysis results, a dynamic privacy protection rule file is generated in combination with big data analysis. The specific steps are as follows:

[0039] The generated encryption analysis results and global model parameters are input into the big data analysis engine to evaluate the performance of the global encryption gradient data;

[0040] The big data analysis engine performs privacy risk analysis based on the encryption analysis results;

[0041] The big data analysis engine generates privacy protection recommendations using rule-based reasoning and machine learning models based on performance evaluation results and privacy risk analysis results;

[0042] Based on the privacy risk analysis results and privacy protection recommendations, a dynamic privacy protection rule file is generated through a decision tree.

[0043] As a preferred solution of the mobile phone information privacy protection method based on big data analysis described in the present invention, the dynamic privacy protection rule file is parsed through the ASM framework to parse the bytecode of the mobile phone application, automatically modify the sensitive API call point, and insert the privacy monitoring logic. The specific steps are as follows:

[0044] Use apktool to decompile the APK and obtain the .dex file;

[0045] Use dex2jar to convert the .dex file into .class bytecode;

[0046] Use the ASM framework to parse the decompiled .class bytecode, and read the .class bytecode through ClassReader to generate the class structure tree of the file;

[0047] By rewriting the visitMethod method, all Java methods are captured and processed in the class structure tree, and the bytecode interceptor of each Java method is generated;

[0048] Combine dynamic privacy protection rule files and bytecode interceptors to automatically identify sensitive API call points in .class bytecodes;

[0049] At the identified sensitive API call points, the bytecode of the privacy monitoring method is inserted before the Java method call through ASM's MethodVisitor, and the result processing logic is inserted after the Java method call.

[0050] As a preferred solution of the mobile phone information privacy protection method based on big data analysis described in the present invention, the modified sensitive API call points are repackaged into an APK file, and the privacy monitoring verification of the application is performed. The specific steps are as follows:

[0051] Use the d8 tool to repackage the modified .class bytecode into a .dex file, replace the original .dex file, and generate a new .dex file;

[0052] Replace the new .dex file into the decompiled file structure and use apktool to repackage and generate a new APK file;

[0053] Use the developer key to sign the new APK file and install it on the smart terminal device;

[0054] Use automated testing tools to verify whether sensitive API call points trigger the inserted privacy monitoring logic;

[0055] If the inserted privacy monitoring logic is triggered, it means that the sensitive API call point is captured by the monitoring logic, relevant log records are generated, and privacy protection measures are executed;

[0056] If the inserted privacy monitoring logic is not triggered, it means that the sensitive API call point is not captured and the log lacks relevant information about the monitoring logic. Check the bytecode, modify and verify the test environment until the inserted privacy monitoring logic is successfully triggered.

[0057] In a second aspect, the present invention provides a mobile phone information privacy protection system based on big data analysis, including an information collection module, a global model parameter generation module, an encryption analysis module, a dynamic privacy protection rule file generation module, a bytecode modification module and a verification module;

[0058] The information collection module is used to collect the latest privacy protection requirements from the application market and / or regulatory authorities, extract privacy protection clauses, and generate a standardized privacy protection rule data set;

[0059] The global model parameter generation module generates global model parameters based on the privacy protection rule data set by using distributed training and encrypted aggregation;

[0060] The encryption analysis module is used to perform encryption analysis on the privacy data of mobile phone applications using multi-party secure computing on global model parameters;

[0061] The dynamic privacy protection rule file generation module is used to generate a dynamic privacy protection rule file in combination with big data analysis after obtaining the encryption analysis results;

[0062] The bytecode modification module is used to parse the bytecode of the mobile application through the ASM framework for the dynamic privacy protection rule file, automatically modify the sensitive API call points, and insert the privacy monitoring logic;

[0063] The verification module is used to repackage the modified sensitive API call points into an APK file and perform privacy monitoring verification on the application.

[0064] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the mobile phone information privacy protection method based on big data analysis as described in the first aspect of the present invention is implemented.

[0065] In a fourth aspect, the present invention provides 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 mobile phone information privacy protection method based on big data analysis as described in the first aspect of the present invention.

[0066] The beneficial effects of the present invention are as follows: by combining technologies such as big data analysis, federated learning and multi-party secure computing, the present invention can realize dynamic and real-time generation and updating of privacy protection rules, overcoming the problems of lag and static nature of privacy protection rules in the prior art; secondly, it can keep up with the latest privacy protection regulations and standards in a timely manner, automatically modify sensitive API call points in applications, and improve the flexibility of privacy protection; in addition, the present invention adopts encryption analysis and distributed training to ensure the secure sharing and processing of privacy data with the participation of multiple parties, effectively enhancing the protection of mobile phone information privacy, and is suitable for complex and changeable application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0068] Figure 1 This is a flow chart of the mobile phone information privacy protection method based on big data analysis in Example 1.

[0069] Figure 2 This is a module diagram of the mobile phone information privacy protection method based on big data analysis in Example 1. DETAILED DESCRIPTION

[0070] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0071] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0072] 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 term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0073] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, and provides a method for protecting mobile phone information privacy based on big data analysis, comprising the following steps:

[0074] S1. Collect the latest privacy protection requirements from the application market and / or regulatory authorities, extract privacy protection clauses, and generate a standardized privacy protection rule dataset. The specific steps are as follows:

[0075] S11. Collect privacy protection requirements from the application market and / or regulatory authorities. The main data sources involved include application market privacy policies and / or regulatory provisions.

[0076] App market privacy policy: Google Play Store, Apple App Store, etc.

[0077] Regulatory requirements: GDPR, CCPA, PIPL (China Personal Information Protection Law) and other global privacy protection regulations.

[0078] Through public API interfaces (such as Google Play Developer API) and web crawlers, automatically obtain the latest privacy protection requirements from application markets and / or regulatory authorities; privacy protection requirements include privacy protection terms and policies.

[0079] For example, suppose there are N application markets and M privacy protection clauses, and the time complexity of data collection through API calls is T api , the expression is: T api =O(N×M).

[0080] It should also be noted that for privacy data that cannot be obtained through the API, a customized crawler system (such as Scrapy) should be used to crawl text information of privacy protection regulations from specific regulatory websites.

[0081] S12. Based on the obtained privacy protection requirements, use NLP methods (such as regular expressions and the natural language processing toolkit SpaCy) to clean up noisy data (HTML tags, special characters, irrelevant text) to ensure data quality.

[0082] For example, assuming that the length of the original input text is T, the noise ratio is p, and the length of the cleaned text is expressed as: T clean =T×(1-p).

[0083] Among them, the cleaned and segmented structured text data contains the privacy regulations of various application markets and regulatory authorities, and is stored in structured JSON or CSV format.

[0084] S13. Use the pre-trained legal domain BERT model (such as Legal-BERT) to perform named entity recognition on the cleaned privacy text and extract privacy protection clauses.

[0085] It should be noted that the BERT model is a pre-trained language model based on the Transformer architecture, which can generate context-sensitive word vectors in natural language processing tasks. The main innovation of BERT is that it is bidirectional, that is, when considering the context of a word, it looks at both the previous word and the following word.

[0086] The core part of the BERT model consists of multiple TransformerEncoder layers, and its basic structure is as follows:

[0087] Input layer: The input sequence is usually a word sequence with special tags. The input format is: [CLS]+sentence 1+[SEP]+sentence 2+[SEP]; [CLS] is a special classification tag used to represent the entire sequence, and [SEP] is used to separate sentences.

[0088] Embedding layer, TokenEmbedding: maps each word into a vector of fixed dimension. SegmentEmbedding: distinguishes different sentences in the input, and is used to distinguish sentence 1 from sentence 2 in the sentence classification task. PositionEmbedding: Since Transformer has no built-in order information, BERT encodes the position of the word in the sentence through position embedding.

[0089] TransformerEncoder layer, each Encoder layer consists of two parts: a multi-head self-attention mechanism and a feedforward neural network, where:

[0090] Multi-Head Self-Attention Mechanism: BERT uses a multi-head self-attention mechanism to allow each word in the context to "pay attention" to all other words before and after it, thereby encoding contextual information.

[0091] Feed-forward Neural Network (FFN): The output after self-attention is transformed nonlinearly through a two-layer feed-forward neural network.

[0092] Different versions of the BERT model contain different numbers of Encoder layers, BERT-base: 12 Encoder layers, hidden layer size is 768, attention heads are 12, and the total parameters are 110M.

[0093] BERT-large: 24 Encoder layers, hidden layer size of 1024, 16 attention heads, and a total of 340M parameters.

[0094] Output layer: In the pre-training task, BERT has two tasks, Masked Language Model (MLM): randomly mask some words in the input and require the model to predict these words.

[0095] Next Sentence Prediction (NSP): The model is required to determine whether two sentences are logically continuous.

[0096] Furthermore, in order to adapt to tasks in the legal field, the BERT model usually needs to be pre-trained on legal corpora for domain adaptation (Domain-Adaptive Pretraining). This is achieved through the following two steps: general BERT pre-training and legal domain adaptive pre-training:

[0097] Specifically, general BERT pre-training: Use the standard pre-training process of BERT, including MLM and NSP tasks, but train on general general corpora (such as Wikipedia, BooksCorpus). This process helps the model learn general language representation.

[0098] Adaptive pre-training in the legal field and collection of domain corpus: In order to adapt the BERT model to the characteristics of legal language, a large amount of legal text needs to be collected. Legal texts include but are not limited to: judgments, regulations, contract clauses, legal papers, privacy policies, etc.

[0099] Continue pre-training: Continue training the BERT model on legal domain corpus, using the same loss function as the original BERT (MLM and NSP tasks). The purpose of this step is to allow the model to better understand legal terminology, context, and norms.

[0100] It should also be noted that the structure of the pre-trained legal BERT model is basically the same as the original BERT model, with the following differences:

[0101] Input structure: The input text mainly comes from legal documents, such as privacy policies, regulations, judgments, etc. The input layer is labeled in the same way as BERT's standard input format (including [CLS] and [SEP] tags).

[0102] Embedding of legal domain corpus: Vocabulary: When pre-training in the legal domain, the model can use the standard BERT vocabulary or generate a new vocabulary based on the legal corpus. The new vocabulary will contain more legal terms and terminology.

[0103] Special legal terms and phrases are embedded into the model’s TokenEmbedding, allowing the model to better process and understand these terms.

[0104] TransformerEncoder layers: The number of Encoder layers is the same as the original BERT model (e.g., BERT-base or BERT-large), but these layers are further trained on legal domain corpora to better capture the contextual relationships in legal language.

[0105] Output layer: The output layer continues to perform MLM and NSP tasks during pre-training, but on text in the legal domain. The goal is to enable the model to have the ability to process legal language.

[0106] Preferably, the pre-trained legal domain BERT model can better understand and process legal documents related to privacy protection by performing domain-adaptive pre-training on legal corpus. In terms of model structure, it is basically the same as the standard BERT model, but through pre-training in the legal field, the model's ability to understand legal language and terminology is enhanced, so that it can more effectively extract key privacy protection information and generate a standardized privacy protection rule data set.

[0107] Among them, the legal field BERT model is used to process and understand legal texts such as privacy protection clauses and policies. The following tasks can be achieved through the pre-trained legal BERT model:

[0108] Named Entity Recognition (NER): Extract key information from the privacy policy, such as legal terms, personal information type, etc.

[0109] Semantic similarity calculation: Match legal terms in different expressions to ensure the consistency of privacy policies and regulations.

[0110] Rule mapping: Map the extracted entities to a standardized privacy protection clause framework to help generate a privacy protection rule dataset.

[0111] S14. According to the privacy protection standards (such as the terms of GDPR and the requirements of CCPA), the extracted privacy protection terms are mapped to the privacy protection terms framework (such as GDPR and CCPA) to unify the format.

[0112] S15. Based on the unified format of the clause framework, use semantic similarity calculation (such as cosine similarity) to match regulatory clauses described in different languages ​​to ensure the consistency of the clauses.

[0113] S16. Standardize regulatory provisions from different sources through rule mapping and hierarchical classification.

[0114] Specifically, based on semantic similarity, regulatory provisions from different sources are mapped into a unified privacy protection standard model, which can be based on GDPR or CCPA.

[0115] The standardized mapping model uses weighted similarity scores to complete the term mapping.

[0116] Classify the clauses into different categories according to different dimensions of privacy protection (such as data collection, data processing, user consent, etc.). Use a hierarchical classification tree for automated classification.

[0117] Output: A standardized privacy protection clause dataset, where each clause is mapped to a specific standard and classified into the corresponding privacy protection dimension.

[0118] S17. Use rule engines and knowledge graphs to verify and check the consistency of standardized regulatory clauses to form a standardized privacy protection rule data set.

[0119] S2, based on the privacy protection rule dataset, uses distributed training and encrypted aggregation to generate global model parameters, including:

[0120] S21. Use a hierarchical federated learning architecture to group privacy-preserving rule datasets into different nodes and layers; each layer represents a different data grouping or cluster of privacy domains. For example, the bottom layer can be a specific application category (such as financial applications, social media applications), the middle layer can be an industry (such as finance, healthcare), and the top layer can be a country or region (such as the European Union, the United States, etc.).

[0121] S22. At each node of each layer, based on the respective data groupings, the model is trained using the local privacy rule data set to obtain the model parameters of each layer.

[0122] S23. Each node calculates the loss function based on the local privacy rule data set, and updates the model parameters of each layer by optimizing the loss function;

[0123] Calculate the loss function L i The expression of (θ) is:

[0124]

[0125] Among them, L i (θ) represents the loss value of the i-th node, θ represents the model parameters, and n i represents the number of data samples of the i-th node, j is the index variable for summation, l represents the loss function, f θ (x i,j ) represents the model f θ For input data x i,j The predicted value, x i,j represents the input feature of the jth data sample in the i-th node, y i,j represents the target value of the jth sample in the i-th node, i represents the index of the node, and j represents the index variable for summation;

[0126] The local privacy rule dataset provides local training data for each node, including input features x i,j and the target value y i,j ;

[0127] The local privacy rule dataset refers to the privacy protection rule subset generated by each node based on its local user data and relevant privacy protection requirements (such as privacy terms, laws and regulations, privacy policies, etc.), which contains the data samples x unique to the node. i,j and the corresponding label y i,j In the expression x i,j and i,j In fact, it is the samples and target values ​​extracted from the local privacy rule dataset of the node. The use of this dataset ensures that the user's sensitive information will not be leaked during the model training process.

[0128] Specifically, the privacy protection rules dataset is a complete privacy protection rules data set generated after cleaning, standardization, mapping and verification of privacy protection requirements, privacy terms and policies collected from application markets, regulatory authorities, etc.

[0129] Local privacy rule dataset: In federated learning or distributed systems, the complete privacy protection rule dataset is distributed to each node. Each node only holds a subset of privacy protection rules related to its local data, which is called the "local privacy rule dataset". This dataset is used for training locally on the node to ensure that data is not directly shared between nodes.

[0130] For example, suppose there are multiple nodes in a federated learning system, and each node holds the privacy data of different users: Global privacy rule dataset: contains the privacy protection terms, privacy policies, data usage rules, etc. of all users, which is summarized from multiple data sources (such as laws, regulations, and privacy policies of different countries).

[0131] Local privacy rule dataset: Each node only uses rules related to its local user privacy data. For example, a node that manages European user data may have its local privacy rule dataset including privacy clauses related to the General Data Protection Regulation (GDPR), while another node that manages Chinese user data may include privacy protection rules related to the Cybersecurity Law.

[0132] The specific “local privacy rule data set” may include:

[0133] Privacy terms: Privacy terms related to the user's private data managed by the node. For example, the user's privacy protection terms for location data, address book, text messages, sensor data, etc.

[0134] Laws and regulations: Privacy laws and regulations related to a specific region, covering requirements for data collection, storage, processing, sharing, etc. For example, if a node manages European user data, its local privacy rules dataset may include the provisions of the GDPR.

[0135] Privacy Policy: The privacy policy formulated by the application market, developer or company specifies how the node handles the user's private data and needs to follow specific privacy protection standards.

[0136] Sensitive data types: Which user data managed by the node is sensitive data (such as GPS location, health data, financial data, etc.), and the protection rules for this data.

[0137] Data processing rules: How to handle user data according to privacy terms and regulations, including requirements for data encryption, storage, and transmission.

[0138] S24, send the updated model parameters of each layer to the central server of the layer through encrypted communication, and use multi-party secure computing (MPC) to encrypt and aggregate the parameters within the layer, and output the global model parameters, which are expressed as:

[0139]

[0140] Among them, θ (t+1) It indicates that after the t+1th round of training, the global model parameters are updated. Indicates the weighted average of the parameters of N nodes, where N represents the total number of nodes. It represents the local model parameters of the i-th node after the t-th round of training, and t represents the number of iterations.

[0141] It should be noted that, first of all, the local model parameters are the model parameter results obtained by each node through local training; the intra-layer parameters are composed of the local model parameters trained by multiple nodes in the same layer, which are sent to the central server of the layer through encrypted communication; the updated model parameters of each layer are the global model parameters obtained after these local model parameters are aggregated through multi-party secure calculation and encryption.

[0142] It should be noted that "multi-party secure computation" is a method for ensuring data privacy, allowing multiple participants to collaborate to complete computing tasks without leaking their own data. In the present invention, multi-party secure computation can be used to achieve secure aggregation of model parameters, that is, without directly exchanging original data, each participant can jointly calculate an updated value of a global model; secondly, multi-party secure computation (MPC) is mainly concerned with protecting data privacy in a single calculation, rather than directly involving multiple rounds of training. However, in the multi-round training process of federated learning, MPC can be applied multiple times to ensure data privacy and security in each round of iteration. Therefore, it can be said that multi-party secure computation supports multi-round training in federated learning, but it is not a multi-round training process itself.

[0143] It should be noted that the expression θ(t+1) This describes a parameter aggregation method in federated learning, where federated learning is essentially an iterative process, and each participant will continuously update their model in multiple rounds until a certain convergence condition is reached. After each round of training, the global model parameters will be updated according to the above formula or other similar methods.

[0144] It should also be noted that the server should use secure multi-party computing (MPC) for parameter aggregation to ensure that no private data is leaked during the aggregation process. At the same time, in the layered architecture, after the lower-level aggregation is completed, the parameters of each layer are submitted upward to the higher-level server until the top-level server completes the final global model aggregation. After the top-level model is aggregated, the global model parameters are generated.

[0145] To ensure the privacy of data during training, group computing and key sharing mechanisms are used for privacy protection in each layer (during training and aggregation of each layer), as expressed in:

[0146]

[0147] Among them, z i represents the encryption result of the i-th node, M represents the total number of generated partial keys, k is the index of the partial key, s i,k represents the kth partial key of the ith node.

[0148] It should be noted that in the aggregation process of each layer, hierarchical MPC is adopted, that is, MPC calculations are first performed inside the nodes of each layer to ensure that the data of this layer will not be leaked, and then the encrypted results are submitted to the upper-level server for further aggregation.

[0149] Preferably, the combination of group computing and hierarchical MPC can effectively reduce the computing pressure of each node and improve the scalability of the overall system. The MPC calculation of each level only needs to process the data of this level, reducing the complexity of cross-layer data transmission.

[0150] A dynamic optimization mechanism is introduced to adjust the global model parameters to adapt to changes in different privacy scenarios.

[0151] By regularly checking the performance of the global model, the model parameters are dynamically adjusted for different privacy scenarios. The expression is:

[0152]

[0153] Among them, η represents the learning rate, Represents the gradient of the global loss function;

[0154] Specifically, nodes at each level can use secure communication to provide feedback to the upper layer about difficulties encountered during training or changes in data distribution. The top-level server dynamically adjusts the training strategy of the global model based on these feedbacks.

[0155] Preferably, group computing and hierarchical MPC greatly reduce the computational burden of each layer, adapting to privacy-preserving training in large-scale distributed scenarios; using MPC technology to perform secure computing between nodes at each level to ensure that data privacy will not be leaked during the entire training process; through a dynamic optimization mechanism, the model can adapt to changes in different privacy scenarios, ensuring the accuracy and flexibility of the global model.

[0156] S3: For global model parameters, multi-party secure computing is used to perform encryption analysis on the privacy data of mobile applications, including:

[0157] S31. Through global model parameters, coordinate each node to perform local training without sharing original data, and update the global model parameters through encrypted aggregation summary nodes.

[0158] S32. The updated global model parameters will be fed back to each node, and each node will collect local privacy data based on the updated global model parameters.

[0159] S33. Standardize local privacy data into a unified feature scale to ensure that data between different nodes can be effectively compared and aggregated to prepare for subsequent encryption and model training.

[0160] S34. Perform homomorphic encryption on the standardized local privacy data to generate an encrypted data set, ensuring that the data remains private during transmission and calculation, and calculations can be performed in an encrypted state.

[0161] The key of each node is sharded, dividing the key into several partial keys, and distributing the sub-keys to multiple parties. The purpose is to increase the security of the data. Even if a sub-key is leaked, the complete key cannot be restored.

[0162] Among them, the original key can only be recovered when a sufficient number of key fragments are combined together, which can ensure the security of the key.

[0163] S35. Each node performs local model training based on its own encrypted data set and calculates encrypted gradient data to prepare for subsequent global model updates.

[0164] S36. The encrypted gradient data is transmitted to the central server, and the encrypted gradient data of each node is encrypted and aggregated through multi-party secure computing (MPC) to obtain global encrypted gradient data, which can ensure that the central server can generate updates to the global model without leaking the specific gradient data of each node.

[0165] It should be noted that through homomorphic encryption and multi-party secure computation (MPC), the aggregation of encrypted gradient data can be completed without decrypting the gradient. MPC allows multiple participants to jointly calculate the value of a function without exposing their own data. Combined with homomorphic encryption, the gradients of each node can be securely aggregated in an encrypted state.

[0166] S37. By updating the gradient descent model parameters based on the global encrypted gradient data and combining the dynamic changes of the time dimension, gradient aggregation and global model update are achieved under full encryption, preventing any single party (including the server) from accessing plaintext data or model parameters, while ensuring that each node can use local data for effective model updates.

[0167] S38. Feedback the updated gradient descent model parameters to each node. Each node uses the subkey to decrypt the global gradient descent model parameters, which can ensure that each node obtains the latest global model parameters and continues local training.

[0168] It should be noted that the purpose of this process is to ensure that the global model parameters remain encrypted during transmission and aggregation, and these parameters can only be obtained after each node decrypts them through the subkey. This mechanism can ensure data privacy and security, and prevent the server or other third parties from snooping on the model parameters during transmission.

[0169] The decrypted global gradient descent model parameters are combined with the encrypted gradient data of each node to generate encrypted analysis results. These analysis results include performance evaluation of the model, contribution analysis of each node, and preliminary assessment of privacy leakage risks. While ensuring the accuracy of the model and the privacy of the data, this analysis result will be used for subsequent big data analysis.

[0170] It should be noted that the encrypted gradient data is the local encrypted gradient from each node; the global encrypted gradient data is the result obtained by aggregating the encrypted gradient data of all nodes.

[0171] The connection between the two is that the global encrypted gradient data is the gradient information from a global perspective obtained by encrypting and aggregating the encrypted gradient data of multiple nodes, which is used to update the global model.

[0172] S4. After obtaining the encryption analysis results, generate a dynamic privacy protection rule file in combination with big data analysis. The specific steps are as follows:

[0173] S41. Input the generated encryption analysis results and global model parameters into the big data analysis engine to evaluate the performance of the global encrypted gradient data, including the convergence of the model, node contribution and training efficiency, as follows:

[0174] The convergence of the model is used to evaluate the convergence speed of the model during training and determine whether there are signs of overfitting or underfitting;

[0175] Node contribution, which is used to analyze the contribution of each node to the model (based on gradient changes) and determine whether there are abnormal nodes or data deviations;

[0176] Training efficiency, used to evaluate the training efficiency and communication latency of nodes and identify network bottlenecks or computing bottlenecks.

[0177] S42. After the performance evaluation, the big data analysis engine conducts a privacy risk analysis based on the encryption analysis results. Specifically, it analyzes the privacy leakage risk, including whether the private data of a certain node can be inferred through the global model; confirms whether the homomorphic encryption algorithm currently used is powerful enough and whether the encryption key length or algorithm needs to be adjusted.

[0178] S43. The big data analysis engine generates privacy protection recommendations based on the performance evaluation results and privacy risk analysis results using rule-based reasoning (RBR) and machine learning models (e.g., supervised learning models). These recommendations include:

[0179] Adjustment of encryption algorithm: whether it is necessary to increase encryption strength or replace encryption algorithm in the next round of training.

[0180] Node-level privacy policy: Generates a personalized privacy protection policy for each node to control potential privacy leakage risks.

[0181] S44. Generate a dynamic privacy protection rule file through automatic reasoning of the decision tree based on the privacy risk analysis results and privacy protection recommendations. The specific process is as follows:

[0182] Train a decision tree model to learn the best privacy-preserving strategy under different privacy risk and performance conditions.

[0183] Input big data analysis results and let the decision tree generate the most appropriate privacy protection strategy.

[0184] Convert the generated policy into a standardized dynamic privacy protection rule file (such as JSON or XML).

[0185] Save and distribute the rule files for use by each training node.

[0186] Among them, the big data analysis results specifically include privacy risk assessment results, privacy compliance analysis reports, system resource consumption and performance evaluation, application behavior detection reports, user privacy preference and behavior analysis results, and feedback and improvement suggestions on privacy protection models.

[0187] Specifically, privacy protection rule files usually refer to a series of files generated by policies and technical measures used to protect data privacy, especially in the fields of distributed systems, federated learning, data processing, etc. The goal of these rules is to ensure that sensitive data is not accessed or leaked without authorization, while maximizing system performance while ensuring privacy.

[0188] Specifically, they include encryption rules: encryption algorithm, key length, encrypted data transmission, etc.; differential privacy rules: enabling differential privacy and parameter settings (such as ε and δ); data processing rules: data sampling, data sharding, anonymization, etc.; security aggregation rules: using secure multi-party computing and security aggregation technology to protect data privacy; access control and permission management: role access control, zero trust architecture, principle of least privilege, etc.; dynamic privacy protection rules: rule update frequency, adaptive privacy protection strategy, etc.; log audit and monitoring rules: privacy event monitoring and audit log management.

[0189] It should also be noted that these rules can be flexibly combined and dynamically adjusted according to different privacy requirements and data scenarios to ensure that system performance is maximized while protecting privacy.

[0190] S5. For the dynamic privacy protection rule file, the ASM framework is used to parse the bytecode of the mobile application, automatically modify the sensitive API call points, and insert the privacy monitoring logic. The specific steps are as follows:

[0191] S51. Decompile the APK using apktool to obtain a .dex file; “APK” stands for Android Package, which is translated into Android application package in Chinese.

[0192] S52. Use dex2jar to convert the .dex file to .class bytecode.

[0193] S53. Use the ASM framework to parse the decompiled .class bytecode, and read each .class bytecode through ClassReader to generate a class structure tree of the file; ClassReader is the core class in the ASM framework for reading .class files, which is responsible for parsing Java bytecodes and passing various parts of the class file (such as fields, methods, etc.) to ClassVisitor for further processing or modification; at the same time, as the starting point of the bytecode modification tool chain, ClassReader is used in conjunction with classes such as ClassVisitor and MethodVisitor to achieve deep operation and modification of Java class files.

[0194] S54. All method calls are captured by overriding the visitMethod method, all Java methods are captured in the class structure tree, and processed to generate a bytecode interceptor for each Java method.

[0195] Among them, visitMethod is an important method in the ASM framework, which is used to capture and process each method definition in a class. Through it, developers can create a MethodVisitor for each method to modify the bytecode instructions inside the method. This is the key entry point for modifying method behavior during the bytecode modification process.

[0196] S55. Combine the dynamic privacy protection rule file and the bytecode interceptor to automatically identify sensitive API call points in the .class bytecode; the rule file defines a list of sensitive APIs that need to be monitored (such as location, camera, contact reading, etc.). During the bytecode parsing process, each method call is compared with the list to identify sensitive APIs that need to be modified.

[0197] S56. At the identified sensitive API call point, insert the bytecode for calling the privacy monitoring method (such as recording call information, user authorization judgment, etc.) before the Java method call instruction through ASM's MethodVisitor, and insert the result processing logic after the Java method call to ensure that the use of sensitive APIs is monitored and restricted in real time; where MethodVisitor is a class name in the ASM framework, dedicated to bytecode operations.

[0198] It should be noted that the specific content of the monitoring logic may include recording the time of API calls, passed parameters, call results, etc., or directly blocking certain calls according to the requirements of the rule file. In the MethodVisitor of ASM, the monitoring code is inserted so that the calling behavior of sensitive APIs can be tracked or restricted.

[0199] S6. Repackage the modified sensitive API call points into an APK file and perform privacy monitoring verification on the application.

[0200] S61. Use the d8 tool to repackage the modified .class file into a .dex file, replace the original .dex file, and generate a new .dex file.

[0201] S62. Replace the new .dex file into the decompiled file structure, and use apktool to repackage and generate a new APK file.

[0202] S63. Use the developer key to sign the new APK file to ensure that it can be installed and run on the Android device and installed on the smart terminal device.

[0203] It should be noted that when signing an APK file, you need to ensure that the correct key and alias are used. If the key is lost or the wrong key is used, the APK will not be installed correctly on the device. At the same time, if you want to publish it to Google Play, you also need to ensure that the signature of the APK is consistent with the previously released version.

[0204] S64. Use automated testing tools (such as Espresso or UIAutomator) to verify whether the call points of sensitive APIs trigger the inserted privacy monitoring logic. If the inserted privacy monitoring logic is triggered, it means that the sensitive API call points are captured by the monitoring logic, relevant log records are generated, and privacy protection measures are executed. If the inserted privacy monitoring logic is not triggered, it means that the sensitive API call points are not captured, and relevant information about the monitoring logic is missing in the log. Check the bytecode, modify and verify the test environment until the inserted privacy monitoring logic is successfully triggered.

[0205] It should be noted that the privacy monitoring logic includes pre-logic and post-logic, as follows:

[0206] Pre-logic: Whether the inserted privacy monitoring method is successfully executed before the call point of the sensitive API is called.

[0207] Post-processing logic: whether the inserted result processing logic is successfully executed after the call point of the sensitive API is called.

[0208] The present embodiment also provides a mobile phone information privacy protection system based on big data analysis, including: an information collection module, a global model parameter generation module, an encryption analysis module, a dynamic privacy protection rule file generation module, a bytecode modification module and a verification module; the information collection module is used to collect the latest privacy protection requirements from the application market and regulatory authorities, and extract key privacy protection information to generate a standardized privacy protection rule data set; the global model parameter generation module is used to generate global model parameters based on the privacy protection rule data set by using distributed training and encryption aggregation; the encryption analysis module is used to perform encryption analysis on the privacy data of mobile phone applications using multi-party secure computing on the global model parameters; the dynamic privacy protection rule file generation module is used to generate a dynamic privacy protection rule file in combination with big data analysis after obtaining the encryption analysis results; the bytecode modification module is used to parse the bytecode of the mobile phone application through the ASM framework for the dynamic privacy protection rule file, automatically modify sensitive API calls and insert privacy monitoring logic; the verification module is used to repackage the modified bytecode into an APK file and perform privacy monitoring verification on the application.

[0209] This embodiment also provides a computer device, which is suitable for the mobile phone information privacy protection method based on big data analysis, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the mobile phone information privacy protection method based on big data analysis proposed in the above embodiment.

[0210] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0211] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for protecting the privacy of mobile phone information based on big data analysis proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, SRAM for short), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, EEPROM for short), erasable programmable read-only memory (Erasable Programmable Read Only Memory, EPROM for short), programmable read-only memory (Programmable Red-Only Memory, PROM for short), read-only memory (Read-Only Memory, ROM for short), magnetic storage, flash memory, disk or optical disk.

[0212] In summary, by combining technologies such as big data analysis, federated learning, and multi-party secure computing, the present invention can realize dynamic and real-time generation and updating of privacy protection rules, overcoming the problems of lag and static nature of privacy protection rules in the prior art; secondly, it can keep up with the latest privacy protection regulations and standards in a timely manner, automatically modify sensitive API call points in applications, and improve the flexibility of privacy protection; in addition, the present invention adopts encryption analysis and distributed training to ensure the secure sharing and processing of privacy data with the participation of multiple parties, effectively enhancing the protection of mobile phone information privacy, and is suitable for complex and changeable application scenarios.

[0213] Example 2, referring to Table 1, is the second example of the present invention. To further verify the technical solution of the present invention, experimental simulation data of a mobile phone information privacy protection method based on big data analysis is provided.

[0214] The experiment selected five applications with high privacy risks, covering social media, instant messaging, financial payment, health data management, and online shopping applications. These applications generally have problems such as excessive collection of sensitive information and unauthorized API calls without any privacy protection treatment. The experiment compared the privacy protection scheme of the prior art with the scheme of the present invention to evaluate their respective privacy protection effects.

[0215] The experiment is divided into two stages: data collection and preprocessing, and comparison of privacy protection schemes:

[0216] Data collection and preprocessing: Use web crawlers and public API interfaces to obtain the latest privacy protection terms and related regulatory information from the application market and regulatory authorities. Use natural language processing (NLP) technology to clean up the noise data in the privacy protection terms, and use the pre-trained BERT model for named entity recognition to extract key information in the privacy protection terms. This step provides a basis for the subsequent generation of privacy protection rules.

[0217] Among them, "named entities" refer to key information or terms with specific legal significance in privacy texts. These entities are usually concepts or nouns involved in privacy policies and legal clauses, and have attributes related to law or privacy protection. Specific named entities can include the following:

[0218] Types of personal information: such as "name", "ID number", "email address", "telephone number", "geolocation data", etc. The privacy protection terms usually clearly list what types of personal information will be collected, processed or shared.

[0219] Name of laws and regulations: The name of laws, regulations or standards related to privacy protection, such as "General Data Protection Regulation (GDPR)", "California Consumer Privacy Act (CCPA)", etc.

[0220] Organization Name: The name of the company, regulatory agency, or third-party organization involved in privacy protection, such as "Google", "Facebook", "National Information Security Administration", etc.

[0221] Data processing activities: specific operations for processing personal data described in the terms, such as "collection", "storage", "sharing", "deletion", "encryption", etc.

[0222] Time and date: refers to the time limit or validity period specified in the privacy protection terms, such as "data retention period", "time limit for data deletion" or "policy effective date".

[0223] Location: The geographical location where data is stored or processed, such as "within the EU", "outside the United States", "within China", etc. These locations may be related to data sovereignty or legal jurisdiction.

[0224] Privacy rights: the rights that users enjoy in the privacy protection terms, such as "right to access", "right to deletion", "right to data portability", "right to object", etc.

[0225] For example, suppose there is a privacy protection clause as follows:

[0226] The Company will collect the user's name, email address and phone number. Under the General Data Protection Regulation (GDPR), users have the right to request that the Company delete their personal data and that the data will only be stored within the EU.

[0227] In this text, named entities can be: name (personal information type), email address (personal information type), phone number (personal information type), General Data Protection Regulation (GDPR) (name of the law and regulation), and EU territory (geographic location).

[0228] These key information (named entities) are a very important part of the privacy policy. They can help understand the specific content and requirements of the privacy protection clauses, and then extract the privacy protection clauses to provide a basis for subsequent rule-making and privacy monitoring.

[0229] Comparison of privacy protection solutions: Existing technical solutions: Privacy protection methods based on static encryption analysis. Existing technical solutions usually perform encryption analysis on the privacy data of applications and define and detect fixed privacy rules at the application level. This solution is characterized by the use of a single server for centralized encryption processing, the lack of dynamic adjustment capabilities for privacy protection rules, and the lack of distributed data processing capabilities. When protecting privacy, the call of sensitive APIs only relies on preset static rules and cannot adapt to the dynamic changes in privacy risks in actual applications.

[0230] The solution of the present invention: a privacy protection solution based on big data analysis and federated learning. The solution first distributes the privacy protection rule data set to multiple nodes through a distributed federated learning architecture, and performs local training on each node. During the training process, multi-party secure computing technology is used to encrypt, transmit and aggregate the model parameters between nodes to ensure that private data will not be leaked during transmission. Subsequently, combined with the homomorphic encryption method, the privacy data of the mobile application is dynamically encrypted and analyzed, and the generated privacy protection rule file can be dynamically adjusted according to the real-time privacy risks. Finally, the application bytecode is automatically modified through the ASM bytecode parsing framework, the privacy monitoring logic is inserted, and the new APK file is repackaged.

[0231] Cryptographic analysis and rule generation: The existing technical solutions use static cryptographic analysis methods, and the privacy protection rules rely on preset static models, and new protection rules cannot be generated dynamically. The solution of the present invention uses a big data analysis engine to conduct privacy risk assessment on the cryptographic analysis results, and generates new privacy protection rules based on dynamic privacy risks.

[0232] Bytecode modification and verification: Both solutions modify the bytecode of the application and insert privacy monitoring logic. The existing technical solution manually marks sensitive APIs and inserts fixed monitoring logic, while the solution of the present invention combines dynamic privacy protection rule files to perform automatic bytecode modification and sensitive API identification.

[0233] Furthermore, in the experiment, the privacy protection effects of unprotected, existing technical solutions and the solution of the present invention were compared. The key indicators of the experiment included: the reduction rate of API sensitive call times, privacy leakage risk score, application performance impact and privacy monitoring trigger rate.

[0234] The details are shown in Table 1 below:

[0235] Table 1 Experimental data comparison table

[0236]

[0237]

[0238] The experimental data comparison results are shown in the table above. The existing technical solutions and the solution of the present invention show significant differences in multiple privacy protection indicators, as follows:

[0239] Reduction in the number of sensitive API calls: In unprotected applications, sensitive APIs are called as many as 120 times. The prior art solution can reduce the number of sensitive API calls to 80 times through static encryption analysis and fixed rule detection, a reduction of about 33%. However, since the prior art solution cannot dynamically adjust the privacy protection rules, the calls to sensitive APIs are still relatively frequent. The solution of the present invention generates dynamic privacy protection rules through big data analysis, and automatically adjusts the API call strategy, further reducing the number of sensitive API calls to 30 times, a reduction of 62.5% compared to the prior art. This shows that the solution of the present invention has significant advantages in real-time monitoring and dynamic adjustment of privacy risks.

[0240] Privacy leakage risk score: The privacy leakage risk score is an important indicator to measure the privacy protection ability of an application. The privacy leakage risk score of an unprotected application is 85 points, indicating that it faces a greater risk of privacy leakage. The existing technical solution reduces the risk score to 50 points through static encryption analysis and detection. Although the risk is reduced, it cannot cope with dynamically changing privacy risks due to the static nature of the rules. The solution of the present invention realizes distributed encryption analysis of privacy data through federated learning and multi-party secure computing technology, and generates dynamic privacy protection rules in combination with big data analysis, further reducing the risk score to 10 points. Compared with the prior art, the solution of the present invention significantly improves the accuracy and flexibility of privacy protection.

[0241] Impact on application performance: The privacy protection schemes of the prior art have a certain impact on performance in actual applications, which is mainly reflected in the decrease in the response speed of the application due to the insertion of encryption analysis and privacy monitoring logic. The prior art scheme has an impact on application performance of 10%, while the scheme of the present invention reduces the performance impact to 5% through optimized encryption calculation and model update mechanism. While protecting privacy, the scheme of the present invention minimizes the negative impact on application performance and ensures that the user experience will not be significantly affected.

[0242] Privacy monitoring trigger rate: The privacy monitoring trigger rate refers to the frequency at which the privacy monitoring logic successfully captures and responds when an application calls a sensitive API. The privacy monitoring trigger rate of the prior art solution is 60%, which can capture calls to sensitive APIs in most cases, but due to the static nature of its rules, there is still a risk that some sensitive API calls are not monitored. The solution of the present invention can capture calls to sensitive APIs in most cases through dynamically generated privacy protection rules, and the monitoring trigger rate is as high as 95%, indicating that the comprehensiveness and real-time nature of its privacy protection have been greatly improved.

[0243] In summary, the mobile phone information privacy protection method based on big data analysis of the present invention is superior to the existing technical solutions in many aspects. Compared with the traditional static encryption analysis method, the scheme of the present invention realizes the dynamic generation and adjustment of privacy protection rules through federated learning and big data analysis technology, which can more effectively reduce the number of sensitive API calls and reduce the risk of privacy leakage, ensuring comprehensive privacy protection of applications without sacrificing user experience.

[0244] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A mobile phone information privacy protection method based on big data analysis, characterized in that: include: Collect the latest privacy protection requirements from the application market and / or regulatory authorities, extract privacy protection clauses, and generate a standardized privacy protection rule dataset; Based on the privacy-preserving rule dataset, distributed training and encrypted aggregation are used to generate global model parameters. For global model parameters, multi-party secure computing is used to encrypt and analyze the privacy data of mobile applications; After obtaining the encryption analysis results, generate a dynamic privacy protection rule file in combination with big data analysis; For dynamic privacy protection rule files, the ASM framework is used to parse the bytecode of mobile applications, automatically modify sensitive API call points, and insert privacy monitoring logic; Repackage the modified sensitive API call points into an APK file and perform privacy monitoring and verification on the application.

2. The method for protecting mobile phone information privacy based on big data analysis as claimed in claim 1, characterized in that: The specific steps of collecting the latest privacy protection requirements from the application market and / or regulatory authorities, extracting privacy protection clauses, and generating a standardized privacy protection rule data set are as follows: Automatically obtain the latest privacy protection requirements from the application market and / or regulatory authorities through public API interfaces and web crawlers; privacy protection requirements include privacy protection terms and policies; Based on the obtained privacy protection requirements, use NLP methods to clean up the noise data in the private text; Use the pre-trained legal domain BERT model to perform named entity recognition on the cleaned privacy text and extract privacy protection clauses; According to the privacy protection standards, the extracted privacy protection clauses are mapped to the clause framework to unify the format; Based on the unified format of the clause framework, semantic similarity is used to match the regulatory clauses described in different languages; Standardize regulatory provisions from different sources through rule mapping and hierarchical classification; Use rule engines and knowledge graphs to verify and check the consistency of standardized regulatory clauses to form a standardized privacy protection rule data set.

3. The method for protecting mobile phone information privacy based on big data analysis as claimed in claim 2, characterized in that: Based on the privacy protection rule data set, distributed training and encrypted aggregation are used to generate global model parameters. The specific steps are as follows: Using a hierarchical federated learning architecture, the privacy-preserving rule datasets are grouped into different nodes and layers; At each node of each layer, based on the respective data grouping, the model is trained using the local privacy rule dataset to obtain the model parameters of each layer; Each node calculates the loss function based on the local privacy rule data set and updates the model parameters of each layer by optimizing the loss function; The updated model parameters of each layer are sent to the central server of the layer through encrypted communication, and multi-party secure computing is used to encrypt and aggregate the parameters within the layer to output the global model parameters. The expression is: Among them, θ (t+1) represents the global model parameters obtained after the t+1th round of training, Indicates the weighted average of the parameters of N nodes, where N represents the total number of nodes. It represents the local model parameters of the i-th node after the t-th round of training, and t represents the number of iterations.

4. The method for protecting mobile phone information privacy based on big data analysis as claimed in claim 3, characterized in that: The global model parameters are encrypted and analyzed by multi-party secure computing on the privacy data of mobile phone applications. The specific steps are as follows: Through global model parameters, coordinate each node to perform local training without sharing original data, and update global model parameters through encrypted aggregation summary nodes; The updated global model parameters will be fed back to each node, and each node will collect local privacy data based on the updated global model parameters; Normalize local privacy data into a unified feature scale; Perform homomorphic encryption on the standardized local privacy data to generate an encrypted data set; Each node performs local model training based on its own encrypted data set and calculates encrypted gradient data; The encrypted gradient data is transmitted to the central server, and the encrypted gradient data of each node is encrypted and aggregated through multi-party secure computing to obtain the global encrypted gradient data; By combining the dynamic changes of the time dimension on the basis of globally encrypted gradient data, the gradient descent model parameters are updated; Feedback the updated gradient descent model parameters to each node, and each node uses the subkey to decrypt the global gradient descent model parameters; Use the decrypted global gradient descent model parameters and combine them with the encrypted gradient data of each node to generate encrypted analysis results.

5. The method for protecting mobile phone information privacy based on big data analysis as claimed in claim 4, characterized in that: After obtaining the encryption analysis results, a dynamic privacy protection rule file is generated in combination with big data analysis. The specific steps are as follows: The generated encryption analysis results and global model parameters are input into the big data analysis engine to evaluate the performance of the global encryption gradient data; The big data analysis engine performs privacy risk analysis based on the encryption analysis results; The big data analysis engine generates privacy protection recommendations using rule-based reasoning and machine learning models based on performance evaluation results and privacy risk analysis results; Based on the privacy risk analysis results and privacy protection recommendations, a dynamic privacy protection rule file is generated through a decision tree.

6. The method for protecting mobile phone information privacy based on big data analysis as claimed in claim 5, characterized in that: The dynamic privacy protection rule file is parsed through the ASM framework to parse the bytecode of the mobile application, automatically modify the sensitive API call points, and insert the privacy monitoring logic. The specific steps are as follows: Use apktool to decompile the APK and obtain the .dex file; Use dex2jar to convert the .dex file into .class bytecode; Use the ASM framework to parse the decompiled .class bytecode, and read the .class bytecode through ClassReader to generate the class structure tree of the file; By rewriting the visitMethod method, all Java methods are captured and processed in the class structure tree, and the bytecode interceptor of each Java method is generated; Combine dynamic privacy protection rule files and bytecode interceptors to automatically identify sensitive API call points in .class bytecodes; At the identified sensitive API call points, the bytecode of the privacy monitoring method is inserted before the Java method call through ASM's MethodVisitor, and the result processing logic is inserted after the Java method call.

7. The method for protecting mobile phone information privacy based on big data analysis as claimed in claim 6, characterized in that: The modified sensitive API call points are repackaged into an APK file, and the application is subjected to privacy monitoring verification. The specific steps are as follows: Use the d8 tool to repackage the modified .class bytecode into a .dex file, replace the original .dex file, and generate a new .dex file; Replace the new .dex file into the decompiled file structure and use apktool to repackage and generate a new APK file; Use the developer key to sign the new APK file and install it on the smart terminal device; Use automated testing tools to verify whether sensitive API call points trigger the inserted privacy monitoring logic; If the inserted privacy monitoring logic is triggered, it means that the sensitive API call point is captured by the monitoring logic, relevant log records are generated, and privacy protection measures are executed; If the inserted privacy monitoring logic is not triggered, it means that the sensitive API call point is not captured and the log lacks relevant information about the monitoring logic. Check the bytecode, and modify and verify the test environment until the inserted privacy monitoring logic is successfully triggered.

8. A mobile phone information privacy protection system based on big data analysis, based on the mobile phone information privacy protection method based on big data analysis according to any one of claims 1 to 7, characterized in that: include: Information collection module, global model parameter generation module, encryption analysis module, dynamic privacy protection rule file generation module, bytecode modification module and verification module; The information collection module is used to collect the latest privacy protection requirements from the application market and / or regulatory authorities, extract privacy protection clauses, and generate a standardized privacy protection rule data set; The global model parameter generation module generates global model parameters based on the privacy protection rule data set by using distributed training and encrypted aggregation; The encryption analysis module is used to perform encryption analysis on the privacy data of mobile phone applications using multi-party secure computing on global model parameters; The dynamic privacy protection rule file generation module is used to generate a dynamic privacy protection rule file in combination with big data analysis after obtaining the encryption analysis results; The bytecode modification module is used to parse the bytecode of the mobile application through the ASM framework for the dynamic privacy protection rule file, automatically modify the sensitive API call points, and insert the privacy monitoring logic; The verification module is used to repackage the modified sensitive API call points into an APK file and perform privacy monitoring verification on the application.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the mobile phone information privacy protection method based on big data analysis described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the mobile phone information privacy protection method based on big data analysis described in any one of claims 1 to 7 are implemented.

Images