Data processing method, system, computing device, and readable storage medium

Abstract

The application discloses a data processing method, system, computing device and readable storage medium. The data processing method is applied to an edge model and comprises the following steps: acquiring first data and at least one processing task of the first data; identifying sensitive data in the first data; extracting second data corresponding to the processing task from the first data; if the second data contains sensitive data, encrypting the second data; and sending the processing task and the second data to a corresponding cloud model, so that the cloud model processes the second data. In this way, the edge model is used to protect data privacy, the cloud model is used to realize efficient data processing, and the safety and processing efficiency of data are ensured.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a data processing method, system, computing device, and readable storage medium. Background Art

[0002] With the development of large-scale language models, artificial intelligence technology is evolving from a tool for solving specific, limited-scope problems to intelligent systems capable of handling a wide range of tasks and possessing higher-level autonomous learning and problem-solving capabilities. However, this process faces a conflict between unlocking data value and ensuring privacy and security, especially when processing sensitive data, where difficult choices must be made between performance and compliance. Current technical solutions face performance bottlenecks in local deployment of edge models and the risk of privacy leakage in cloud-based models. There is an urgent need to address the challenge of achieving dynamic privacy protection while ensuring data availability. Summary of the Invention

[0003] The purpose of this application is to provide a data processing method, a computing device, and a computer-readable storage medium, which utilize edge models to protect data privacy and achieve efficient data processing through cloud models, thereby ensuring data security and processing efficiency.

[0004] To achieve the above objectives:

[0005] In a first aspect, an embodiment of the present application provides a data processing method, comprising the following steps:

[0006] Acquiring first data and at least one processing task for the first data;

[0007] identifying sensitive data in the first data;

[0008] extracting second data corresponding to the processing task from the first data;

[0009] If sensitive data exists in the second data, encrypting the second data;

[0010] The processing task and the second data are sent to the corresponding cloud model, so that the cloud model processes the second data.

[0011] In one embodiment, before extracting the second data corresponding to the processing task from the first data, the method includes:

[0012] Split the processing task into multiple processing subtasks;

[0013] Extracting second data corresponding to the processing task from the first data includes:

[0014] Second data corresponding to each processing subtask is extracted from the first data.

[0015] In one embodiment, identifying sensitive data in the first data includes:

[0016] executing a corresponding identification strategy according to the structure type of the first data to identify sensitive data in the first data;

[0017] Encrypting the second data includes:

[0018] Determine the corresponding encryption strategy based on the type of sensitive data;

[0019] Encrypt sensitive data based on encryption policies.

[0020] In one embodiment, the processing tasks are split, including:

[0021] Processing tasks are split according to the type of sensitive data and / or semantic information of the processing tasks.

[0022] In one embodiment, the type of sensitive data includes at least one of personnel information, user data, financial transaction data, business secrets, marketing strategies, management decisions, and security keys.

[0023] In one embodiment, sending the processing task and the second data to the corresponding cloud model includes:

[0024] Determining a corresponding cloud model according to the processing task and the type of the second data;

[0025] The processing task and the second data are sent to the target interface, so that the cloud model connected to the target interface processes the second data according to the processing task.

[0026] In one embodiment, after sending the processing task and the second data to the corresponding cloud model, the method includes:

[0027] Receiving third data returned by the cloud model after processing the second data;

[0028] If there is encrypted information in the third data, decrypting the encrypted information;

[0029] After aggregating the third data, the processing result is output.

[0030] In a second aspect, an embodiment of the present application provides a data processing system, including an edge model and a cloud model;

[0031] The edge model is configured to obtain first data and at least one processing task for the first data; identify sensitive data in the first data; extract second data corresponding to the processing task from the first data; encrypt the second data if sensitive data is present in the second data; and send the processing task and the second data to a corresponding cloud model, causing the cloud model to process the second data.

[0032] The cloud model is used to receive and process the second data.

[0033] In a third aspect, an embodiment of the present application provides a computing device, specifically comprising:

[0034] processor;

[0035] a memory for storing processor-executable instructions;

[0036] The processor is configured to execute instructions for executing the data processing method as described in the first aspect.

[0037] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which a computer program is stored. When the instructions in the computer-readable storage medium are executed by a processor of a computing device, the computing device is enabled to implement the data processing method described in the first aspect.

[0038] The data processing method, system, computing device, and readable storage medium provided in the embodiments of the present application are applied to an edge model and include: obtaining first data and at least one processing task for the first data; identifying sensitive data in the first data; extracting second data corresponding to the processing task from the first data; encrypting the second data if sensitive data is present in the second data; and sending the processing task and the second data to a corresponding cloud model, causing the cloud model to process the second data. In this way, data privacy is protected by the edge model and efficient data processing is achieved through the cloud model, ensuring data security and processing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A flowchart of a data processing method provided by an embodiment of the present invention.

[0040] Figure 2 A schematic diagram of a specific flow chart of the data processing method provided in an embodiment of the present invention.

[0041] Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] Exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with the present application. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present application, as detailed in the appended claims.

[0043] It should be noted that, in this document, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element. In addition, components, features, and elements with the same name in different embodiments of the present application may have the same meaning or different meanings, and their specific meanings need to be determined by their explanation in the specific embodiment or further combined with the context of the specific embodiment.

[0044] It should be understood that although the terms first, second, third, etc. may be used herein to describe various information, such information should not be limited to these terms. These terms are merely used to distinguish information of the same type from one another. For example, without departing from the scope of this document, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the term "if" as used herein may be interpreted as "at the time of," "when," or "in response to a determination." Furthermore, as used herein, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context indicates otherwise. It should be further understood that the terms "comprising" and "including" indicate the presence of the described features, steps, operations, elements, components, items, types, and / or groups, but do not exclude the presence, occurrence, or addition of one or more other features, steps, operations, elements, components, items, types, and / or groups. The terms "or" and "and / or" as used herein are to be interpreted as inclusive, meaning any one or any combination. Thus, “A, B, or C” or “A, B, and / or C” means “any of: A; B; C; A and B; A and C; B and C; A, B, and C.” An exception to this definition occurs only when a combination of elements, functions, steps, or operations are inherently mutually exclusive in some manner.

[0045] It should be understood that, although the various steps in the flowchart in the embodiment of the present application are shown in sequence according to the indication of the arrows, these steps are not necessarily performed in sequence in the order indicated by the arrows. Unless clearly stated herein, the execution of these steps is not strictly limited in order, and they can be performed in other orders. Moreover, at least a portion of the steps in the figure may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and their execution order is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0046] It should be noted that in this article, step codes such as S1 and S2 are used for the purpose of expressing the corresponding content more clearly and concisely, and do not constitute a substantial restriction on the order. When implementing the step, those skilled in the art may execute S2 first and then S1, etc., but these should all be within the scope of protection of this application.

[0047] It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application.

[0048] In the subsequent description, the use of suffixes such as "module", "component" or "unit" to represent elements is only for the purpose of facilitating the description of the present application and has no specific meaning. Therefore, "module", "component" or "unit" can be used interchangeably.

[0049] Existing technologies allow for the local deployment of large language models as edge models. Open-source models (such as Llama3-70B) can be privately deployed, but this requires expensive hardware, such as an 8xA100 GPU cluster, and inference latency reaches sub-seconds, making it difficult to meet the demands of real-time interaction. For example, in text generation, open-source models perform approximately 34% lower than GPT-4 in terms of intent understanding accuracy and coherence, significantly lagging behind in performance.

[0050] Directly invoking large language models deployed in the cloud, known as cloud-based models, poses a risk of privacy breaches. Directly transmitting raw data to third-party APIs violates regulations such as GDPR and HIPAA. An IBM Security report indicates that 23% of healthcare data breaches in 2023 were caused by API vulnerabilities. Attackers can infer sensitive inputs from model outputs. For example, a Google study revealed that 90% of the original training data can be reconstructed with just 50 API queries ("Extracting Training Data from LLMs," 2023).

[0051] First embodiment

[0052] See Figure 1To address the above challenges, an embodiment of the present application provides a data processing method. In this embodiment, the data processing method is applied to a computing device as an example, where an edge model is deployed on the computing device. The data processing method provided in this embodiment includes:

[0053] Step S1: Acquire first data and at least one processing task for the first data.

[0054] The edge model receives first data, which may come from user input, sensors, database query results, or information captured by a web crawler. It also determines at least one processing task for the first data, which may include the specific content of the processing task. The processing task is completed by performing operations such as filtering, analyzing, and calculating the first data to extract valid information.

[0055] It should be noted that the edge model refers to a lightweight artificial intelligence model deployed on a terminal device or local server, and the cloud model refers to a high-performance language processing model running in a remote data center, providing strong semantic understanding and complex computing capabilities through APIs.

[0056] Step S2: Identify sensitive data in the first data.

[0057] Sensitive data refers to information that contains personal privacy, business secrets, or is protected by regulations. This includes, but is not limited to, personnel information, user data, financial transaction data, trade secrets, marketing strategies, management decisions, and security keys. By identifying and recording sensitive data, or outputting the identification results in a structured format (e.g., JSON that annotates the location, type, and confidence level of sensitive fields), objects can be provided for data encryption operations.

[0058] Specifically, personnel information includes at least personal information (such as name, ID number, contact information, salary details, performance appraisals, and labor contracts) and personnel data (such as interview evaluations, background check reports, and information on unhired personnel). User data includes at least identity data (such as name, address, phone number, and bank card number), behavioral data (such as consumption records, app usage history, and search keywords), and service interaction data (such as customer complaint records, customer service recordings, and sensitive details in repair work orders). Financial transaction data includes at least core financial data (such as income / expense details, profit margins, tax returns, and investment plans) and sensitive transaction information (such as bank account passwords, M&A negotiation floor prices, and supplier contract prices). Trade secrets include at least R&D secrets (such as undisclosed patents, product design drawings, and production process parameters) and digital assets (such as core algorithms, source code, private datasets, and data analysis models). Market strategy includes at least strategic planning (such as business plans, market expansion roadmaps, and product launch timelines) and competitive intelligence (such as pricing strategies, bid quotations, and internal competitive analysis reports). Management decisions should at least include high-level decisions (such as board resolutions, equity distribution plans, and management meeting recordings) and audit and compliance records (such as internal audit reports and responses to regulatory inquiries). Security keys should at least include security protection information (such as firewall configurations, vulnerability scan reports, and intrusion detection logs) and key encryption information (such as SSL certificate private keys, database encryption keys, and API access tokens).

[0059] Step S3: extracting second data corresponding to the processing task from the first data.

[0060] According to the content of the processing task, such as specific business needs or analysis objectives, the required second data is filtered out from the first data. The extraction process includes but is not limited to the splitting or reorganization of the first data. Specifically, the feasibility analysis of the decomposition of the processing task is performed through the edge model. Based on factors such as task complexity, the first data type and the processing capability of the cloud model, it can be determined whether the processing task needs to be split. If the processing task needs to be split, the processing task will be split before subsequent encryption and calling of cloud models are performed. If splitting is not required, encryption and calling of cloud models are performed directly.

[0061] Step S4: If the second data contains sensitive data, encrypt the second data.

[0062] By associating the first and second data, it can be determined whether sensitive data identified in the first data is present in the second data. Sensitive data can also be identified again on the second data to avoid missing sensitive data that appears in the second data after data splitting or reassembly.

[0063] If sensitive data exists in the second data, the second data is encrypted so that the second data only includes encrypted data and / or non-sensitive data. Encryption is enabled only for sensitive data, and non-sensitive fields are transmitted directly to avoid computational burden. First, sensitive data, such as personal identity information, financial records, or health information, is identified and marked. Next, the corresponding encryption strategy, such as encryption algorithm and key management, is determined, which can be achieved by pre-defining a mapping relationship between sensitive data types and encryption strategies. Then, the corresponding encryption strategy is applied to encrypt the sensitive data, and the encrypted data set is saved as the second data, thereby effectively preventing the leakage of sensitive data.

[0064] Step S5: Send the processing task and the second data to the corresponding cloud model, so that the cloud model processes the second data.

[0065] The second data and processing tasks are sent to the corresponding cloud models. Different cloud models are pre-evaluated based on their processing expertise (e.g., GPT-4 supports natural language generation, Vertex AI is adaptable to encrypted numerical computation, and Claude 3 handles medical advice generation), data processing type, and processing performance. Cloud models are then labeled or categorized based on the evaluation results. Different types of data and their associated processing tasks are then dynamically matched to different cloud models for processing. For example, encrypted payment data can be routed to Vertex AI, desensitized diagnostic text can be sent to Claude 3, and text analysis tasks can be sent to GPT-4. This enables secure and efficient task distribution and cloud-to-cloud collaboration, improving data processing efficiency.

[0066] In this way, the embodiment of the present application uses the above-mentioned dynamic data encryption and analysis method based on edge model-cloud model collaboration to deploy a lightweight model locally (edge), such as a 500MB model, to complete steps such as sensitive data identification and task splitting, and then call the cloud model to complete more complex data analysis and task processing steps, thereby avoiding privacy leakage and effectively reducing hardware costs.

[0067] In one embodiment, before extracting the second data corresponding to the processing task from the first data, the method includes:

[0068] Split the processing task into multiple processing subtasks;

[0069] Extracting second data corresponding to the processing task from the first data includes:

[0070] Second data corresponding to each processing subtask is extracted from the first data.

[0071] Optionally, if it is determined that the processing task needs to be split, it can be split based on the type of sensitive data and / or the semantic information of the processing task. The natural language understanding module of the edge model can be used to parse the processing task, such as "analyze patient health risks and generate payment reports", and split it into logically independent subtasks, such as "health risk assessment" and "payment trend statistics". Dependencies between the processing subtasks are also generated, such as completing the diagnostic analysis before generating financial advice. In addition, different processing subtasks can be generated based on different types of sensitive data to ensure that a single processing subtask does not expose the entire privacy chain.

[0072] Then, according to the different processing subtasks, the required fields are extracted from the first data, and the contextual semantic fragments of the field are retained through association analysis (such as dependency syntax parsing) (such as "medication records" need to be associated with "diagnosis results" to support rationality analysis) to form the second data.

[0073] In one embodiment, identifying sensitive data in the first data includes:

[0074] A corresponding identification strategy is executed according to the structural type of the first data to identify sensitive data in the first data.

[0075] Optionally, the first data is scanned to determine the structure type of each field in the first data, including structured data (such as records in a database table), semi-structured data (such as JSON or XML format), or unstructured data (such as free text, images, etc.).

[0076] For structured data, you can match column names or labels based on predefined sensitive field dictionaries (such as "ID number" and "bank card number"), and use regular expressions to verify the data format. Set thresholds for fields such as amount and date (for example, a single payment exceeding 100,000 yuan triggers a high-risk mark) to identify sensitive behaviors that are potentially associated with outliers. For example, in a customer information table, fields such as "name", "address", and "credit card number" may contain sensitive data. For semi-structured data, you can use XPath or JSONPath to locate sensitive nodes and find sensitive data based on tag names or attribute values. For example, search for specific tags in an XML file (such as <creditcardnumber>) to locate sensitive data. For unstructured data: Natural language processing techniques are used to analyze text content and identify sensitive keywords or phrases. For example, trained edge models can be used to detect personally identifiable information (PII) in text, or optical character recognition (OCR) can be used to extract text from images and then assess sensitive data.

[0077] In one embodiment, encrypting the second data includes:

[0078] Determine the corresponding encryption strategy based on the type of sensitive data;

[0079] Encrypt sensitive data based on encryption policies.

[0080] Optionally, for known sensitive data types, corresponding encryption strategies, such as encryption algorithms and key management methods, are pre-matched. By using differentiated encryption strategies, privacy can be guaranteed while reducing computational time.

[0081] For example, for personnel information, an encryption strategy of sharding, desensitizing, and encrypting metadata anchors can be adopted. First, the data is sharded, then the metadata is encrypted using the AES-GCM algorithm, and the shard ID and double hash (SHA-256 + SHA3-512) are recorded to ensure the integrity and security of the data during transmission and storage.

[0082] For user data, a selective encryption + dynamic noise mapping encryption strategy can be employed. Sensitive fields are protected through Paillier homomorphic encryption, while non-sensitive fields are dynamically noised to confuse potential attackers. Noise parameters are also encrypted and stored in metadata. Once the cloud-based model returns the results, homomorphic decryption is performed and the noise is reversed for calibration.

[0083] For financial transaction data, a homomorphic noise + automated calibration encryption strategy can be used. Paillier is used to encrypt transaction amounts, while metadata is used to record the encrypted public key fingerprint. After the model returns the encrypted result, it is directly decrypted using the local private key, facilitating subsequent decryption operations without the need to record noise parameters.

[0084] For commercial secrets, a sharding + multi-KMS key encryption strategy can be adopted. Data is split into multiple fragments and encrypted separately. Each fragment uses a different key management service (such as AWS KMS and Google Cloud KMS). The shard key ID is annotated with metadata. API calls require multi-KMS joint authorization and decryption to ensure high security.

[0085] For market strategies, an encryption strategy of contextual obfuscation + AI denoising method can be adopted. Unrelated paragraphs generated by AI are inserted into the text for obfuscation, and metadata is used to mark the location information of the noise paragraphs. After the results are returned, these synthetic contents are filtered through the local edge model.

[0086] For management decisions, you can adopt the encryption strategy of JWT token + temporary storage server, generate a JWT token with a timestamp and IP whitelist for each access, temporarily store encrypted data on a private server, set automatic expiration and release, use the API to access the temporary data with the token, and decrypt the result locally after returning it to ensure that only authorized users can access the encrypted data.

[0087] For security keys, an encryption strategy of double hash + digital signature can be adopted. After data encryption, double hash (SHA-256 + SHA3-512) is calculated, and an RSA digital signature is attached to the metadata. The consistency of the hash value and signature is verified before decryption to prevent data tampering.

[0088] Furthermore, for other undefined sensitive data types, custom encryption strategies can be extended based on standardized interfaces to ensure universality and flexibility in data processing. Users can leverage open, standard interfaces to develop targeted processing methods based on specific business logic and security requirements. Developers can introduce new algorithms, adjust data processing flows, or add additional security measures, and seamlessly integrate them into existing systems through standard interfaces. This approach not only enhances the system's adaptability, enabling it to handle diverse application scenarios, but also maintains the consistency and stability of the architecture, ensuring efficient collaboration between different modules and improving overall performance.

[0089] In one embodiment, sending the processing task and the second data to the corresponding cloud model includes:

[0090] Determining a corresponding cloud model according to the processing task and the type of the second data;

[0091] The processing task and the second data are sent to the target interface (ie, API), so that the cloud model connected to the target interface processes the second data according to the processing task.

[0092] Optionally, cloud APIs are dynamically matched based on one or more of the second data's encryption type, task semantic tags (such as "statistical mean analysis"), and the cloud API's real-time load. The processing task and the second data are then routed to the cloud API, such as routing encrypted payment data to Vertex AI and sending desensitized diagnostic text to Claude 3. Through dynamic routing decisions, high-computational load tasks are assigned to the cloud model, while the local edge model only processes lightweight logic, reducing hardware deployment costs. Furthermore, dynamic API selection based on semantic analysis, data type, and real-time cloud load avoids the logical rigidity caused by manually preset sharding rules.

[0093] In one embodiment, after sending the processing task and the second data to the corresponding cloud model, the method includes:

[0094] Receiving third data returned by the cloud model after processing the second data;

[0095] If there is encrypted information in the third data, decrypting the encrypted information;

[0096] After aggregating the third data, the processing result is output.

[0097] Optionally, upon receiving the third data, the system checks whether it contains encrypted information. If so, it decrypts the encrypted portion using the corresponding key and decryption algorithm (e.g., the decryption method corresponding to AES-GCM or Paillier encryption) according to the previously defined encryption policy. After decryption, if multiple third data items are returned by multiple cloud-based models, they are aggregated, including but not limited to data merging, formatting unification, or result reorganization, to integrate the scattered data fragments into a complete and coherent set of information.

[0098] Next, the aggregated third data is checked to see if it meets the requirements of the initial processing task. This includes, but is not limited to, data integrity, accuracy, compliance, and consistency. For numerical processing results, integrity checking ensures that all expected numerical data is present. For example, in the payment analysis subtask, all monthly payment records (e.g., encrypted amounts of 1500, 1800, and 2000 yuan) are calculated. Accuracy checking verifies that the returned values ​​meet expectations. For example, when calculating patient out-of-pocket payment patterns, the decrypted monthly mean and variance are reasonable, and any logical errors are detected when compared to the original input data. Compliance checking verifies whether the values ​​adhere to specific regulations or standards. For example, whether medical expense statistics comply with medical insurance policies or financial regulations. Consistency checking verifies that the numerical results are consistent with the intended target. For example, whether the calculated average monthly out-of-pocket payment for myocardial infarction patients is consistent with clinical reality and can provide a reference for subsequent treatment.

[0099] For text processing results, a completeness check ensures that all necessary information is included. For example, when generating personalized health recommendations, key elements such as dietary advice, exercise recommendations, and follow-up requirements are covered. An accuracy check verifies the accuracy of the text content, including the use of medical terminology and the rationality of the recommendations. For example, is the statement "Aspirin (inhibits platelet aggregation) and nitroglycerin (dilates coronary arteries) are standard treatments for myocardial infarction" accurate? A compliance check verifies whether the text content complies with relevant laws, regulations, and industry standards. For example, whether health recommendations are based on the latest medical guidelines and whether patient privacy is considered. A match check confirms whether the text content matches the initial requirements. For example, based on the patient's specific situation (such as a diagnosis of myocardial infarction and current use of aspirin and nitroglycerin), whether the rehabilitation recommendations provided are targeted and effective in helping the patient recover.

[0100] If the aggregated results do not meet the requirements of the processing task, steps S1 through S3 of the data processing method are repeated until the results meet the processing task. Alternatively, if some data in the aggregated data fails (e.g., decryption failure of a shard), the corresponding processing subtask can be retriggered for only the problematic shard, avoiding repetition of the entire process. Alternatively, for tasks that fail continuously (e.g., after three reprocessing attempts, the task still fails to meet the requirements), manual intervention is initiated, and the results of the exception handling are recorded.

[0101] Finally, the results that have been aggregated and passed the inspection are output as the final task processing results, thus realizing the full process management from data sending, cloud processing to result acquisition, which not only ensures the security of data during transmission and processing, but also ensures the accuracy and availability of the processing results.

[0102] See Figure 2 The specific steps of the data processing method of the embodiment of the present application are as follows:

[0103] First, the user enters the original requirement and text, which is then intelligently analyzed by the edge model. Next, a feasibility analysis of task decomposition is performed to determine whether task splitting is necessary. If splitting is not necessary, the process proceeds directly to global sensitive data identification, dynamic encryption policy selection and execution, and then an API call is made. If splitting is necessary, task splitting and routing decisions are made, breaking the task into multiple subtasks (such as Subtask 1, Subtask 2, Subtask N, etc.). Each subtask undergoes preprocessing, sensitive data scanning, and dynamic encryption policy selection, and finally calls the corresponding API (such as API-1, API-2, API-N, etc.). After all subtasks are processed, the final response is output through multi-level local aggregation of results.

[0104] For example, a hospital needs to analyze a patient's electronic health record (EHR), which contains the following sensitive data: personal information (such as the patient's name, ID number, contact information, etc.), medical data (such as diagnosis reports, medication records, surgical records, etc.), and financial information (such as medical insurance account numbers, out-of-pocket payment records, etc.). The first data is as follows:

[0105] "Analysis of cardiovascular disease patient data in 2023:

[0106] Patient Li Si (ID number: 310XXXXXX, medical insurance account number: SH123456) was diagnosed with myocardial infarction, and medication records included aspirin and nitroglycerin.

[0107] Self-funded payment records in 2023: 1,500 yuan in January, 1,800 yuan in February, and 2,000 yuan in March.

[0108] The processing task is to use the online APIs of different large model companies to complete analysis tasks and generate disease trend reports, while ensuring that sensitive data is sharded and encrypted and that each API only processes part of the incomplete information.

[0109] First, sensitive data was identified, including personal information (name, ID number) and financial information (medical insurance account number, payment amount). Next, a feasibility analysis was conducted on the task, which was then broken down into three subtasks: analyzing the correlation between diagnosis and medication, analyzing patient payment patterns, and generating personalized health recommendations. These subtasks were performed by online APIs from various leading model companies, such as OpenAI GPT-4, Google Vertex AI, and Anthropic Claude 3. Each API receives data with sensitive information removed or encrypted, as follows:

[0110] Subtask 1 (OpenAI GPT-4):

[0111] Requirement: Analyze the rationality of medication in patients with myocardial infarction.

[0112] Enter the text: "Medication records for myocardial infarction patients: aspirin, nitroglycerin." (Remove personal information).

[0113] Subtask 2 (Google Vertex AI):

[0114] Requirements: Calculate monthly averages and fluctuation trends based on encrypted payment data.

[0115] Input data: Encrypted payment record (Encrypted (1500), Encrypted (1800), Encrypted (2000)).

[0116] Subtask 3 (Anthropic Claude 3):

[0117] Requirement: Generate patient rehabilitation recommendations.

[0118] Input text: "An anonymous patient (ID-P001) was diagnosed with myocardial infarction and is currently taking aspirin and nitroglycerin.

[0119] Next, perform subtask preprocessing and API calls. In the subtask 1 processing channel, call GPT-4. Through sensitive data scanning, confirm that fields such as ID card number and medical insurance account number have been removed. Determine the encryption strategy: No encryption is required, send directly to GPT-4. The API request content is as follows:

[0120] {

[0121] "prompt":"Based on the medication records of 'aspirin and nitroglycerin', analyze whether the treatment plan for patients with myocardial infarction is reasonable. Pharmacological evidence must be included.",

[0122] "max_tokens":300

[0123] }

[0124] After GPT-4 processing, the result returned was "Rational drug combination: Aspirin (inhibits platelet aggregation) and nitroglycerin (dilates coronary arteries) are the standard treatment for myocardial infarction. Be aware of aspirin's gastrointestinal side effects."

[0125] In the subtask 2 processing channel, Vertex AI is called. The payment amount is detected through sensitive data scanning. Encryption strategy: Use Paillier homomorphic encryption for payment data, such as Encrypt (1500). The API request content is as follows:

[0126] {

[0127] "instances":[

[0128] {"encrypted_value":"X1Y2Z3"}, / / encryption (1500)

[0129] {"encrypted_value":"A4B5C6"}, / / encryption (1800)

[0130] {"encrypted_value":"D7E8F9"} / / Encryption (2000)

[0131] ],

[0132] "task":"calculate_mean_and_variance"

[0133] }

[0134] After Vertex AI processing, the returned results are as follows:

[0135] {

[0136] "encrypted_mean":"K0L1M2",

[0137] "encrypted_variance":"N3O4P5"

[0138] }

[0139] In the subtask 3 processing channel, Claude 3 is called. The sensitive data scan determines the anonymized patient ID (ID-P001). Encryption strategy: No encryption is required, and the data is sent directly to the API. The API request content is as follows:

[0140] {

[0141] "prompt":"Generate rehabilitation recommendations for patients with myocardial infarction (medications: aspirin, nitroglycerin), including lifestyle modifications and follow-up requirements.",

[0142] "max_tokens":200

[0143] }

[0144] After Claude 3's processing, the result returned was "Rehabilitation recommendations: low-fat diet, 30 minutes of walking daily; weekly blood pressure monitoring; repeat cardiac ultrasound after 3 months."

[0145] After API processing, the results of each subtask are decrypted and verified through local aggregation, integrated into the final analysis report, and sensitive information in the output content is desensitized. Specifically, the return result of subtask 2 is decrypted and the encrypted result is decrypted using the local private key as follows:

[0146] Monthly average: Decryption("K0L1M2") → 1766.67 yuan

[0147] Variance: Decrypted("N3O4P5") → 58333.33

[0148] Perform data association on subtask 3 and restore ID-P001 to the real medical insurance account number SH123456 through the local mapping table. Perform a logical check on the results of subtask 2 to verify that the hash value of the decrypted result of subtask 2 matches the original data to prevent tampering. Finally, integrate the results of the three APIs and supplement the desensitized personal information as follows:

[0149] "Patient Li Si (ID number: 310***XXX, medical insurance account number: SH12****)

[0150] Diagnosis and medication: Myocardial infarction. Current medication is reasonable (aspirin + nitroglycerin), and attention should be paid to gastrointestinal side effects.

[0151] Payment analysis: The average monthly out-of-pocket payment in 2023 is 1,766.67 yuan (significant fluctuation, variance 58,333.33).

[0152] Rehabilitation recommendations: low-fat diet, daily exercise, and regular echocardiograms."

[0153] Before output, ID numbers and medical insurance account numbers are partially masked (e.g., 310***XXX, SH12****). The final output response includes desensitized personal information, disease diagnosis and medication rationality analysis, out-of-pocket payment analysis, and rehabilitation advice. This entire process ensures the security of sensitive data while achieving the goal of medical data analysis.

[0154] In summary, the data processing method provided in the above embodiment obtains first data and at least one processing task for the first data; identifies sensitive data in the first data; extracts second data corresponding to the processing task from the first data; encrypts the second data if sensitive data is present; and sends the processing task and second data to the corresponding cloud model, which then processes the second data. This approach utilizes edge models to protect data privacy and enables efficient data processing through cloud models, ensuring both data security and processing efficiency.

[0155] Second embodiment

[0156] Based on the same inventive concept as the aforementioned embodiment, an embodiment of the present invention provides a data processing system, including an edge model and a cloud model;

[0157] The edge model is configured to obtain first data and at least one processing task for the first data; identify sensitive data in the first data; extract second data corresponding to the processing task from the first data; encrypt the second data if sensitive data is present in the second data; and send the processing task and the second data to a corresponding cloud model, causing the cloud model to process the second data.

[0158] The cloud model is used to receive and process the second data.

[0159] In one embodiment, before extracting the second data corresponding to the processing task from the first data, the method includes:

[0160] Split the processing task into multiple processing subtasks;

[0161] Extracting second data corresponding to the processing task from the first data includes:

[0162] Second data corresponding to each processing subtask is extracted from the first data.

[0163] In one embodiment, identifying sensitive data in the first data includes:

[0164] executing a corresponding identification strategy according to the structure type of the first data to identify sensitive data in the first data;

[0165] Encrypting the second data includes:

[0166] Determine the corresponding encryption strategy based on the type of sensitive data;

[0167] Encrypt sensitive data based on encryption policies.

[0168] In one embodiment, the processing tasks are split, including:

[0169] Processing tasks are split according to the type of sensitive data and / or semantic information of the processing tasks.

[0170] In one embodiment, the type of sensitive data includes at least one of personnel information, user data, financial transaction data, business secrets, marketing strategies, management decisions, and security keys.

[0171] In one embodiment, sending the processing task and the second data to the corresponding cloud model includes:

[0172] Determining a corresponding cloud model according to the processing task and the type of the second data;

[0173] The processing task and the second data are sent to the target interface, so that the cloud model connected to the target interface processes the second data according to the processing task.

[0174] In one embodiment, after sending the processing task and the second data to the corresponding cloud model, the method includes:

[0175] Receiving third data returned by the cloud model after processing the second data;

[0176] If there is encrypted information in the third data, decrypting the encrypted information;

[0177] After aggregating the third data, the processing result is output.

[0178] For the specific implementation of the data processing system in this embodiment, please refer to the relevant description of the data processing method in the first embodiment, which will not be repeated here.

[0179] Based on the same inventive concept as the above embodiments, an embodiment of the present invention provides a computing device, such as Figure 3 As shown, the computing device includes: a processor 410 and a memory 411 storing a computer program; wherein, Figure 3 The processor 410 shown in the figure is not used to indicate that the number of processors 410 is one, but is only used to indicate the positional relationship of the processor 410 relative to other devices. In actual applications, the number of processors 410 may be one or more; similarly, Figure 3 The memory 411 shown in FIG has the same meaning, that is, it is only used to refer to the position relationship of the memory 411 relative to other devices. In actual applications, the number of memories 411 can be one or more. When the processor 410 runs the computer program, the above data processing method is implemented.

[0180] The computing device may also include: at least one network interface 412. The various components in the computing device are coupled together via a bus system 413. It is understood that the bus system 413 is used to achieve connection and communication between these components. In addition to the data bus, the bus system 413 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, Figure 3 Various buses are labeled as bus system 413.

[0181] Memory 411 may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Non-volatile memory may include read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), ferromagnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disk, or compact disc read-only memory (CD-ROM); magnetic surface memory may include magnetic disk or tape memory. Volatile memory may include random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), synchronous static random access memory (SSRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct rambus random access memory (DRRAM).The memory 411 described in the embodiments of the present invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0182] The memory 411 in the embodiment of the present invention is used to store various types of data to support the operation of the computing device. Examples of such data include: any computer programs used to operate on the computing device, such as operating systems and applications; contact data; phone book data; messages; images; videos, etc. The operating system includes various system programs, such as a framework layer, a core library layer, and a driver layer, which are used to implement various basic services and handle hardware-based tasks. Application programs can include various application programs, such as media players and browsers, which are used to implement various application services. Here, the program implementing the method of the embodiment of the present invention can be included in the application program.

[0183] Based on the same inventive concept as the above-mentioned embodiment, this embodiment further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program. The computer-readable storage medium may be a magnetic random access memory (FRAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a flash memory, a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM); or various devices including one or any combination of the above-mentioned memories, such as a mobile phone, a computer, a tablet device, a personal digital assistant, etc. When the computer program stored in the computer-readable storage medium is executed by a processor, the data processing method applied to the above-mentioned computing device is implemented. For the specific steps implemented when the computer program is executed by the processor, please refer to Figure 1 The description of the illustrated embodiment will not be repeated here.

[0184] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0185] As used herein, the terms "comprises," "includes" or any other variations thereof are intended to cover a non-exclusive inclusion of elements other than the listed elements and may also include additional elements not specifically listed.

[0186] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.< / creditcardnumber>

Claims

1. A data processing method, characterized in that: Applied to edge models, it includes the following steps: Acquire first data and at least one processing task for the first data; identifying sensitive data in the first data; extracting second data corresponding to the processing task from the first data; If the second data contains sensitive data, encrypting the second data; sending the processing task and the second data to a corresponding cloud model, so that the cloud model processes the second data; Before extracting the second data corresponding to the processing task from the first data, the method includes: Splitting the processing task according to the type of the sensitive data and / or the semantic information of the processing task to obtain a plurality of processing subtasks; The extracting second data corresponding to the processing task from the first data includes: extracting second data corresponding to each of the processing subtasks from the first data respectively; The identifying of sensitive data in the first data includes: executing a corresponding identification strategy according to the structure type of the first data to identify sensitive data in the first data; The encrypting the second data includes: Determining a corresponding encryption strategy according to the type of the sensitive data; encrypting the sensitive data based on the encryption policy; The sending the processing task and the second data to the corresponding cloud model includes: Determining a corresponding cloud model according to the processing task and the type of the second data; The processing task and the second data are sent to a target interface, so that the cloud model connected to the target interface processes the second data according to the processing task.

2. The method according to claim 1, characterized in that The type of sensitive data includes at least one of personnel information, user data, financial transaction data, business secrets, market strategies, management decisions and security keys.

3. The method according to claim 1, characterized in that After sending the processing task and the second data to the corresponding cloud model, the method includes: Receiving third data returned by the cloud model after processing the second data; If there is encrypted information in the third data, decrypting the encrypted information; After aggregating the third data, a processing result is output.

4. A data processing system, characterized in that: Including edge models and cloud models; The edge model is configured to obtain first data and at least one processing task for the first data; identify sensitive data in the first data; extract second data corresponding to the processing task from the first data; encrypt the second data if the sensitive data exists in the second data; and send the processing task and the second data to a corresponding cloud model so that the cloud model processes the second data. The cloud model is configured to receive and process the second data; Before extracting the second data corresponding to the processing task from the first data, the method includes: Splitting the processing task according to the type of the sensitive data and / or the semantic information of the processing task to obtain a plurality of processing subtasks; The extracting second data corresponding to the processing task from the first data includes: extracting second data corresponding to each of the processing subtasks from the first data respectively; The identifying of sensitive data in the first data includes: executing a corresponding identification strategy according to the structure type of the first data to identify sensitive data in the first data; The encrypting the second data includes: Determining a corresponding encryption strategy according to the type of the sensitive data; encrypting the sensitive data based on the encryption policy; The sending the processing task and the second data to the corresponding cloud model includes: Determining a corresponding cloud model according to the processing task and the type of the second data; The processing task and the second data are sent to a target interface, so that the cloud model connected to the target interface processes the second data according to the processing task.

5. A computing device, characterized in that include: A processor and a memory for storing executable instructions; wherein the processor is configured to execute the instructions to implement the data processing method according to any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that When the instructions in the computer-readable storage medium are executed by a processor, the data processing method according to any one of claims 1 to 3 is implemented.

Images