A data processing method and apparatus

Abstract

The application discloses a data processing method and device, and relates to the technical field of computers. The method comprises the following steps: in response to monitoring the start of an external large model, intercepting input content to be sent by a user to the external large model, performing content sensitivity detection on the input content, and identifying sensitive information in the input content; calculating a sensitivity level of the input content according to the sensitive information, performing data security processing on the sensitive information in the input content according to the sensitivity level; sending the safe input content after the data security processing to the external large model, obtaining an inference result output by the external large model, and generating a response result according to the inference result and the input content. The embodiment realizes accurate, fine-grained and non-perceptual data processing, performs fine-grained data security processing on the sensitive information based on the sensitivity level, sends the safe input content to the external large model, and generates a response result that is imperceptible to the user based on the inference result output by the external large model and the original input content.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a data processing method and apparatus. Background Technology

[0002] When using external large-scale models for business interactions, data is transmitted over the network to third-party model service platforms, posing a risk of leakage of business data, user privacy, and internal confidential information. Performing content sensitivity detection on the input content before submitting it to the external large-scale model can proactively identify various sensitive information within the input content, thereby enhancing interaction security.

[0003] In the process of realizing this invention, the inventors discovered the following problems in the prior art: Existing sensitivity detection methods typically use simple rule matching and uniformly block requests, resulting in low detection accuracy and an overly simplistic approach. They lack fine-grained security processing of input content, leading to the complete blocking of any input content containing sensitive information. This results in the direct rejection of many input requests that contain a few sensitive fields but have normal business value, impacting the user experience. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide a data processing method and apparatus that achieves precise, fine-grained, and seamless data processing. Based on sensitivity levels, it performs fine-grained data security processing on sensitive information in the input content, sends the secure input content after data security processing to an external large model, and finally generates a response result that is imperceptible to the user based on the inference result output by the external large model and the original input content.

[0005] To achieve the aforementioned objective, according to one aspect of the present invention, a data processing method is provided, comprising: In response to the detection of the startup of an external large model, the system intercepts the user's input content to be sent to the external large model, performs content sensitivity detection on the input content, and identifies sensitive information in the input content. Based on the sensitive information, calculate the sensitivity level of the input content, and perform data security processing on the sensitive information in the input content according to the sensitivity level; The secure input content, after data security processing, is sent to the external large model. The inference result generated by the external large model based on the secure input content is obtained. Based on the inference result and the input content, a response result corresponding to the input content is generated.

[0006] Optionally, content sensitivity detection is performed on the input content to identify sensitive information in the input content, including: combining the historical input sent by the user to the external large model within a specified time with the input content to obtain a combined input; performing rule matching sensitivity detection on the combined input, and if no sensitive information is detected by the rule matching detection, performing semantic sensitivity detection on the combined input to identify explicit sensitive information in the input content; using a local detection model to simulate the reasoning chain of the external large model on the combined input, and detecting the sensitivity of the reasoning chain to identify implicit sensitive information in the input content; and determining the sensitive information in the input content based on the explicit and implicit sensitive information.

[0007] Optionally, calculating the sensitivity level of the input content based on the sensitive information includes: identifying the content type of the input content, where the content type includes at least one of code type, configuration type, log type, and natural language type; calculating the sensitivity of the sensitive information based on the sensitive information, the frequency of occurrence of the sensitive information, and the timeliness of the sensitive information; and calculating the sensitivity level of the input content based on the sensitivity, the content type, and the user's role characteristics.

[0008] Optionally, generating a response result corresponding to the input content based on the reasoning result and the input content includes: identifying content to be determined in the reasoning result, wherein the content to be determined requires the input content to participate in the determination; obtaining the content to be determined according to the reasoning logic corresponding to the content to be determined based on the input content; and assembling the obtained content to be determined and the determined content in the reasoning result to generate the response result.

[0009] Optionally, based on the sensitivity level, data security processing is performed on the sensitive information in the input content, including: if the sensitivity level is level one, replacing or generalizing the sensitive information; if the sensitivity level is level two, partially covering the sensitive information, or adjusting the expression of the input content containing the sensitive information, and sending a corresponding processing permission request to the user to obtain the user's permission.

[0010] Optionally, the method further includes: if the sensitivity level exceeds level two, intercepting the sending action of the input content and prompting the user that there is highly sensitive information in the input content.

[0011] Optionally, the method further includes: in response to receiving updated feedback information about sensitive information, updating the detection criteria for content sensitivity detection and / or the calculation parameters for the sensitivity level based on the updated feedback information.

[0012] According to a second aspect of the present invention, a data processing apparatus is provided, comprising: The sensitivity detection module is used to intercept the user's input content to be sent to the external large model in response to the detection of the external large model's startup, perform content sensitivity detection on the input content, and identify sensitive information in the input content; The first processing module is used to calculate the sensitivity level of the input content based on the sensitive information, and to perform data security processing on the sensitive information in the input content based on the sensitivity level. The second processing module is used to send the secure input content after data security processing to the external large model, obtain the inference result generated by the external large model based on the secure input content, and generate the response result corresponding to the input content according to the inference result and the input content.

[0013] According to a third aspect of the present invention, a data processing electronic device is provided, comprising: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method provided in the first aspect of the embodiments of the present invention.

[0014] According to a fourth aspect of the present invention, a computer-readable medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method provided in the first aspect of the present invention.

[0015] According to a fifth aspect of the present invention, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the method provided in the first aspect of the present invention.

[0016] One embodiment of the invention has the following advantages or beneficial effects: by intercepting the user's input content to be sent to the external large model in response to the detection of the external large model's startup, performing content sensitivity detection on the input content, and identifying sensitive information in the input content; calculating the sensitivity level of the input content based on the sensitive information, and performing data security processing on the sensitive information in the input content based on the sensitivity level; sending the data security-processed secure input content to the external large model, obtaining the inference result generated by the external large model based on the secure input content, and generating the response result corresponding to the input content based on the inference result and the input content, the technical solution achieves accurate, fine-grained, and seamless data processing. Based on the sensitivity level, fine-grained data security processing is performed on the sensitive information in the input content, and the data security-processed secure input content is sent to the external large model. Finally, based on the inference result output by the external large model and the original input content, a response result that is imperceptible to the user is generated. Attached Figure Description

[0017] The accompanying drawings are provided to better understand the invention and are not intended to unduly limit the scope of the invention. Wherein: Figure 1 This is a schematic diagram of the main flow of the data processing method according to an embodiment of the present invention; Figure 2 This is a flowchart of a data processing method according to a possible embodiment of the present invention; Figure 3 This is a flowchart of a data processing method according to another possible embodiment of the present invention; Figure 4 This is a detailed flowchart illustrating the data processing method according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the main modules of a data processing apparatus according to an embodiment of the present invention; Figure 6 This is an exemplary system architecture diagram in which embodiments of the present invention can be applied; Figure 7 This is a schematic diagram of the structure of a computer system suitable for implementing terminal devices or servers of the present invention. Detailed Implementation

[0018] It should be noted that the collection, updating, analysis, use, transmission, and storage of user personal information involved in the technical solution of this invention all comply with relevant laws and regulations, are used for legitimate and reasonable purposes, and are not shared, disclosed, or sold outside of these legitimate uses, and are subject to supervision and management by national regulatory authorities. Necessary measures should be taken to selectively block the use or access to personal information data to prevent unauthorized access to such personal information data, ensure that personnel authorized to access personal information data comply with relevant laws and regulations, and ensure the security of user personal information. Furthermore, once this user personal information data is no longer needed, the risk should be minimized by restricting or even prohibiting data collection and / or deleting the data.

[0019] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of the present invention, including various details to aid understanding. These details should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0020] Existing sensitivity detection methods typically use simple rule matching and uniformly block requests, resulting in low detection accuracy and an overly simplistic approach. They lack fine-grained security processing of input content, leading to the complete blocking of any input content containing sensitive information. This results in the direct rejection of many input requests that contain a few sensitive fields but have normal business value, impacting the user experience.

[0021] To address the aforementioned problems in existing technologies, this invention proposes a data processing method. Based on the identified sensitive information in the input content, a sensitivity level is calculated. According to the sensitivity level, the sensitive information in the input content undergoes data security processing. The security-processed input content is then sent to an external large-scale model. Finally, based on the inference results generated by the external large-scale model and the original input content, a final response result is generated. This achieves precise, fine-grained, and seamless data processing. Specifically, based on the sensitivity level, fine-grained data security processing is performed on the sensitive information in the input content, and the security-processed input content is sent to an external large-scale model. Finally, based on the inference results output by the external large-scale model and the original input content, a response result that is imperceptible to the user is generated.

[0022] Figure 1 This is a schematic diagram of the main flow of the data processing method according to an embodiment of the present invention, such as... Figure 1 As shown, the data processing method of this embodiment includes the following steps S101 to S103.

[0023] Step S101: In response to the detection of the startup of the external large model, the user's input content to be sent to the external large model is intercepted, the input content is subjected to content sensitivity detection, and sensitive information in the input content is identified.

[0024] When users send interaction requests to third-party external large models through internal business systems, given the open operating environment and data storage of these external large models, which are not under the control of the enterprise, it will be difficult to effectively control and recall the interaction text if it involves sensitive information of the enterprise. Therefore, it is highly necessary to intercept the input content before the user sends it to the external large model and to perform sensitive information detection on the intercepted input content.

[0025] The internal business system monitors whether users initiate operations to leave the enterprise's controllable environment. Specifically, it can pre-maintain a list of controlled environments, a list of trusted access terminals, and a whitelist of internal network addresses. The monitoring unit compares the current operation context and the access source environment identifier in real time. In this embodiment of the invention, the monitoring unit is deployed in the enterprise's integrated development environment (IDE), enterprise proxy gateway, AIP forwarding gateway, and browser plugins. If the system detects that a user has launched an external large model in these environments, it proactively intercepts the input content entered by the user in the external large model dialog box, or extracts the prompt word from the request message and uses the prompt word as the input content.

[0026] Understandably, the present invention monitors the startup of the external large model and intercepts the input content to be sent, both of which are performed in the user's normal operation process. This is achieved through event interception and transparent sampling, without requiring any additional operation from the user. The complete input content to be sent is captured without the user's awareness.

[0027] Furthermore, after intercepting and capturing the input content to be sent, the system converts the input content to UTF-8 format, processes special characters such as newlines and tabs, removes leading and trailing whitespace, and normalizes the input content. Next, it performs content sensitivity detection on the normalized input content to check whether it contains sensitive internal information of the enterprise. If sensitive information is involved, it is identified. The sensitive information in this embodiment of the invention may include personal privacy information such as employee or customer ID card numbers and bank card numbers, or it may include core internal business data of the enterprise, such as product costs and contract amounts. If the input content is code, it may also include keys, network tokens, application programming interface (API) keys, etc.

[0028] Step S102: Calculate the sensitivity level of the input content based on the sensitive information, and perform data security processing on the sensitive information in the input content based on the sensitivity level.

[0029] Specifically, the system determines the weight corresponding to the specific semantics of the sensitive information identified from the input content, uses weighted calculation to calculate the sensitivity of the sensitive information, and maps the calculated sensitivity to the corresponding sensitivity level. Understandably, there can be multiple pieces of sensitive information identified from the input content. In the case of multiple pieces of sensitive information, each piece of sensitive information and its corresponding weight are weighted and summed, and the result of the weighted sum is used as the sensitivity of the input content.

[0030] Furthermore, based on the sensitivity level determined by the sensitivity mapping, the specific processing method for sensitive information in the input content is determined. When the sensitivity level is low, it indicates that the sensitive information in the input content is not core content, and data security processing can be performed by replacing it with a dedicated identifier, such as replacing the sensitive information or replacing part of the sensitive information. When the sensitivity level is high, a more intelligent adjustment of the expression of the input content containing sensitive information can be used for data security processing, such as adjusting the input content from calculating departmental indicator A to calculating the general calculation formula of indicator A.

[0031] This invention calculates the sensitivity of input content based on weights determined by the semantics of sensitive information, and then maps these weights to corresponding sensitivity levels. Different data security processing strategies are then employed to process the data according to the sensitivity level, achieving precise and fine-grained data security processing.

[0032] Step S103: Send the secure input content after data security processing to the external large model, obtain the inference result generated by the external large model based on the secure input content, and generate the response result corresponding to the input content according to the inference result and the input content.

[0033] The system replaces the original input content with secure input content generated after the aforementioned data security processing, and sends the secure input content to an external large model. The external large model performs inference based on the received secure input content and outputs the inference result for the secure input content.

[0034] Considering that the inference results output by the external large model are based on secure input content, which differs from the user's original input content, this difference may lead to a discrepancy between the inference results output by the external large model and the user's expected response. To ensure that the user receives an accurate response that matches the original input content, the system typically needs to perform secondary processing on the inference results output by the external large model locally.

[0035] Specifically, after receiving the inference results output by the external large model, the system's local secondary processing unit determines the corresponding processing strategy based on the differences between the secure input content and the user's original input content. This strategy could involve field restoration, where the special identifiers used in the inference results are restored to sensitive information. For example, if the external large model outputs an invitation containing the meeting name "XXX," and since the meeting name is business-sensitive information, data security processing is performed. The local secondary processing unit will restore the fields related to the meeting name "XXX" in the invitation text output by the external large model using the actual meeting name input by the user, generating a response result that is more suitable to the input content and imperceptible to the user.

[0036] Understandably, if the system determines that it needs to prompt the user to be aware that sensitive information has been used, it can attach relevant prompts to the response, such as informing the user that sensitive information was involved when using an external large model. This allows the user to directly use / call the response generated by the local secondary processing unit while simultaneously serving as a reminder of the proper use of sensitive information.

[0037] According to one embodiment of the present invention, the method further includes: in response to receiving updated feedback information about sensitive information, updating the detection criteria for content sensitivity detection and / or the calculation parameters for the sensitivity level based on the updated feedback information.

[0038] Considering that existing sensitive information processing lacks a user feedback learning mechanism, which hinders the system's learning and optimization of sensitive data security processing strategies, this embodiment of the invention introduces an update feedback mechanism to optimize the overall performance of the system's data security processing. The system listens for update feedback information regarding sensitive information from users. This update feedback information can be a system misjudgment feedback when the user disagrees with the sensitive information prompt upon receiving the response result, or it can be sensitive information update information proactively initiated by the user based on easily overlooked sensitive information encountered in daily work.

[0039] Based on the received update feedback, the system updates its detection criteria for content sensitivity detection of input content. For example, it updates the matching rules for sensitive information, adds matching rules for easily overlooked sensitive information to the regular expression matching rules, and can also adjust the weight of sensitive information when calculating the sensitivity level as needed to improve the accuracy of the system's content sensitivity detection and sensitivity level calculation for the next input content to be sent.

[0040] This invention addresses user input content (including text, configuration, code, etc.) to an external large-scale model. By detecting and identifying sensitive information and quantifying its corresponding sensitivity level, it employs a fine-grained, graded processing method to perform data security processing on sensitive information, enhancing the system's intelligence in handling sensitive information. Instead of a simple blocking method, it sends the processed, secure input content to the external large-scale model and, based on the original input content, performs secondary processing on the model's inference results to generate a response that more closely matches the original input content. This achieves seamless sensitive information processing for the user.

[0041] Figure 2 This is a flowchart of a data processing method according to a possible embodiment of the present invention. As another embodiment of the present invention, such as... Figure 2 As shown, the data processing method may include: Step S201: In response to detecting the startup of the external large model, intercept the user's input content to be sent to the external large model.

[0042] Step S202: Combine the historical inputs that the user has sent to the external large model within a specified time with the input content to obtain combined input.

[0043] Considering that user interactions with external large-scale models are multi-round, the currently intercepted input content may have some correlation with the user's historical input. For example, if a user continues to ask follow-up questions about a topic they interacted with yesterday, simply detecting the current input content may affect the accuracy of detection due to the lack of semantic context. For instance, if the user's current input contains an address, it's unknown whether this address is a publicly available company address or a private address. However, if the user previously mentioned a recommendation for address information in a PPT presentation for a trade show, and the current input also mentions trade shows, then based on historical input, the system can determine that the address is a publicly available company address and not sensitive information.

[0044] Therefore, to further improve the accuracy of detecting whether the input content involves sensitive information, this embodiment of the invention combines the user's historical input within a recent period to perform sensitivity detection on the current input content. Specifically, it combines the user's historical input sent to an external large model within a specified time period with the current input content to obtain a combined input.

[0045] Step S203: Perform rule matching sensitivity detection on the combined input, and if no sensitive information is detected by the rule matching detection, perform semantic sensitivity detection on the combined input to identify explicit sensitive information in the input content.

[0046] The system uses combined inputs containing various content as the detection target. It first performs rule-based sensitivity testing on these combined inputs. Considering that the combined inputs can be various types of content such as text and code, the system includes rules applicable to various types of content. Specifically, based on a regular expression library, regular expression matching rules are constructed to match information such as mobile phone numbers, ID cards, bank card numbers, email addresses, IP addresses, and timestamps contained in the text. Based on the key verification bit generation algorithm in the code, corresponding verification bit generation rules are used to match keys, network tokens, and API keys involved in the code. Simultaneously, for proprietary terms within the enterprise, such as project codes, product names, internal domain names, and customer names, this embodiment of the invention uses a trie algorithm or a multi-pattern string matching AC automaton algorithm, employing corresponding dictionary matching rules to quickly detect and match proprietary terms within the enterprise.

[0047] By performing sensitivity detection on combined inputs using the three rule-matching methods described above, the most obvious sensitive information in the combined inputs can be detected quickly and efficiently (in milliseconds). If no sensitive information is detected by rule-matching sensitivity detection, it is possible that there is no sensitive information in the input content, or that the sensitive information in the input content is not so obvious. In order to ensure detection accuracy, this embodiment of the invention also needs to perform semantic-level detection (in seconds) on the combined inputs for which no sensitive information was detected.

[0048] Considering that the combined input may be text, code, or both, the system utilizes a lightweight pre-trained language model (such as BERT or DistilBERT) for text semantic sensitivity detection. It performs binary or multi-label classification on the text in the input content to determine whether the text belongs to high-risk documents such as bidding contracts, technical solutions, user privacy data, or API documentation. Simultaneously, it uses an Abstract Syntax Tree (AST) for code semantic sensitivity detection. Lexical and syntactic analysis is performed on the code fragments in the combined input to generate an AST. Variable names, function names, class names, string constants, and comment content are extracted from the AST, and these elements are then checked for sensitive patterns or the presence of sensitive words such as "password" or "secret."

[0049] Step S204: Use a local detection model to simulate the reasoning thought chain of the external large model on the combined input, and detect the sensitivity of the reasoning thought chain to identify implicit sensitive information in the input content.

[0050] Understandably, both rapid rule-matching sensitivity detection and deep detection of textual and code semantic sensitivity are aimed at identifying explicit sensitive information in the input content, such as keys in code or technical solutions in text. However, in actual interaction, implicit sensitive information may also exist. For example, in a combined input, the input might involve calculating the percentage of department A's profit relative to the company's total profit. This input is a percentage, not a specific amount, and the aforementioned semantic detection alone might fail to detect the risk. When calculating this percentage, the external large model needs to consider both department A's profit and the company's total profit, both of which are core sensitive information.

[0051] Therefore, in this embodiment of the invention, a lightweight local detection model is also deployed locally on the system. This detection model is used to simulate the reasoning chain of an external large model for a combination of inputs containing input content. The reasoning chain is a series of steps followed by the model in reasoning its input task, that is, breaking down the final problem into several sub-problems with sequential execution relationships. By solving these sub-problems, the final problem answer is obtained.

[0052] The local detection model breaks down the problem to be solved from the combined input, obtaining a reasoning chain composed of various sub-problems. By detecting whether solving each sub-problem involves sensitive information, it identifies implicit sensitive information in the input content. Taking the example above, the local detection model analyzes "the percentage of department A's profit in the company's total profit." Calculating the percentage requires obtaining both department A's profit and the company's total profit. Solving these two sub-problems involves the company's core sensitive information. Therefore, based on the reasoning chain of the local detection model, the system identifies "the percentage of department A's profit in the company's total profit" as implicit sensitive information.

[0053] Step S205: Determine the sensitive information in the input content based on the explicit sensitive information and the implicit sensitive information.

[0054] Specifically, the explicit and implicit sensitive information identified above are used as the sensitive information of the input content. That is, the sensitive information in this embodiment of the invention includes explicit and implicit sensitive information.

[0055] Step S206: Identify the content type of the input content, the content type including at least one of code type, configuration type, log type, and natural language type; calculate the sensitivity of the sensitive information based on the sensitive information, the frequency of occurrence of the sensitive information, and the timeliness of the sensitive information; calculate the sensitivity level of the input content based on the sensitivity, the content type, and the user's role characteristics.

[0056] Based on the sensitive information identified above, the system will count the number of times the sensitive information appears in the input content, the timeliness of the sensitive information (such as technical information of products about to be launched), and use a sensitivity mapping method to weight and fuse the corresponding semantic sensitivity, frequency sensitivity, and timeliness sensitivity according to the semantics, frequency sensitivity, and timeliness sensitivity of the sensitive information to obtain a multi-dimensional comprehensive sensitivity of the sensitive information.

[0057] Furthermore, given that the same sensitive information carries different risks across different content types—for example, the input content containing the same sensitive information might be in a test sample and a production environment, with the production environment posing a significantly higher risk than the test sample—this embodiment of the invention also identifies the specific content type of the input content and uses heuristic rules to determine the content type of the requested content in order to accurately assess the sensitivity level of the sensitive information in the input content.

[0058] Specifically, the input content can be identified by its file extension, whether it contains common keywords for code, configuration, and logs (e.g., code keywords: function, def, class, public static void; configuration keywords: host, port, password; log keywords: timestamp patterns [INFO], [ERROR], [DEBUG], [WARNING]), and whether the number of keywords exceeds the corresponding threshold (e.g., 2 code keywords, 3 configuration keywords, and 2 log keywords). If the characteristics of the input content do not meet the above rules, it indicates that the input content is natural language text.

[0059] Of course, if the input content is mixed, it can be segmented according to the content delimiters (such as \n) in the input content, and the content type of each segmented request sub-content can be identified by keywords.

[0060] Furthermore, the user's role and identity also play an important role in assessing the sensitivity level of input content. For example, the sensitivity level of input content is more affected by core R&D personnel and outsourced personnel than by ordinary R&D personnel.

[0061] Therefore, this embodiment of the invention comprehensively considers the sensitivity Wi of the input content, the coefficient Ct corresponding to the content type of the input content, and the coefficient Cu corresponding to the user's role characteristics to calculate the sensitivity level R of the input content. The specific calculation formula is as follows: R = (ΣWi)×Ct×Cu; The sensitivity accumulation operation represents the sensitivity of each piece of sensitive information contained in the input content. Ct and Cu can be pre-set with coefficients corresponding to various content types and user role characteristics. For example: production environment configuration type Ct=1.5, code type Ct=1.3, log type Ct=1.2, contracts, tenders, business plans Ct=1.4, data tables, customer lists Ct=1.4, general explanatory text Ct=1.0, sample code, test cases, teaching templates Ct=0.6. Core R&D users, architects, operations administrators, database administrators Cu = 1.2, ordinary R&D, testing personnel, product managers Cu = 1.0, HR, administrative, finance, sales users Cu = 1.1, outsourced personnel, temporary account users, low-trust users Cu = 1.3.

[0062] Regarding the sensitivity of sensitive information, in addition to determining it through the multi-dimensional comprehensive calculation mentioned above, this embodiment of the invention also provides a faster and more convenient mapping determination method. Based on prior knowledge, this embodiment of the invention predefines the sensitivity references corresponding to various sensitive information in common scenarios, and the sensitivity Wi can be quickly determined by looking up the table.

[0063] For example: Wi = 100 for plaintext private keys, SSH private keys, and certificate private keys; Wi = 90 for Access Keys, Secret Keys, Tokens, JWTs, and API Keys; Wi = 85 for database account passwords, database connection strings, and cached connection strings; Wi = 85 for VPN credentials, bastion host accounts, and maintenance passwords; Wi = 80 for ID card numbers, bank card numbers, and customer real-name information; Wi = 40 for internal IP addresses, internal domain names, service names, and interface paths; Wi = 60 for core SQL snippets and data table design snippets; Wi = 70 for core business code and algorithm implementation snippets; Wi = 75 for bid prices, contract amounts, and undisclosed business terms; and Wi = 30 for general identifiers such as employee email addresses, mobile phone numbers, and employee ID numbers.

[0064] Step S207: If the sensitivity level is Level 1, replace or generalize the sensitive information.

[0065] If the sensitivity level of the input content calculated above is within the threshold range corresponding to Level 1, then the sensitivity level of the input content is Level 1. In this case, the sensitive information in the input content is not core content, and data security processing of the input content can be performed simply by replacing or generalizing the sensitive information.

[0066] For example, replace the phone number in the input with 138 Alternatively, the specific customer name of the enterprise can be generically replaced with "Customer B". Of course, token mapping can also be used to generate a corresponding token locally for sensitive information that needs to maintain a unique correspondence (such as an ID card number), and use the token to replace the ID card number. The mapping relationship between the ID card number and the token is stored locally.

[0067] Step S208: When the sensitivity level is level two, partially cover the sensitive information, or adjust the expression of the input content containing the sensitive information, and send a corresponding processing permission request to the user to obtain the user's permission.

[0068] If the sensitivity level of the input content calculated above is higher than level one, and the sensitive information in the input content belongs to core content, the system determines the sensitivity level to be level two. Since level two is considered a high sensitivity level, this embodiment of the invention will partially mask the sensitive information or adjust the way the input content containing sensitive information is expressed. For example, if the user inputs "calculate the quarterly profit of department A," the system will adjust the input content to "method for calculating quarterly profit." formula.

[0069] In addition, considering that the second-level sensitivity level allows for a large adjustment of sensitive information or input content, it is preferable to send a corresponding processing permission request to the user. Only after the user grants permission for the processing will the processed secure input content be effective.

[0070] Step S209: Send the secure input content after data security processing to the external large model, obtain the inference result generated by the external large model based on the secure input content, and generate the response result corresponding to the input content according to the inference result and the input content.

[0071] The embodiments of the present invention perform precise, multi-dimensional, and hierarchical calculations and processing for content sensitivity detection, sensitivity level calculation, and data security processing, realizing intelligent and secure control of user interaction with external large models, ensuring the security of internal information while allowing users to use external large models.

[0072] Figure 3 This is a flowchart of a data processing method according to another possible embodiment of the present invention. As yet another embodiment of the present invention, such as... Figure 3 As shown, the data processing method may include: Step S301: In response to the detection of the startup of the external large model, the user's input content to be sent to the external large model is intercepted, the input content is subjected to content sensitivity detection, and sensitive information in the input content is identified.

[0073] Step S302: Calculate the sensitivity level of the input content based on the sensitive information. If the sensitivity level is Level 1, replace or generalize the sensitive information. If the sensitivity level is Level 2, partially cover the sensitive information or adjust the expression of the input content containing the sensitive information, and send a corresponding processing permission request to the user to obtain the user's permission.

[0074] Step S303: If the sensitivity level exceeds level two, intercept the sending action of the input content and prompt the user that there is highly sensitive information in the input content.

[0075] To further enhance the system's data security processing performance, this embodiment of the invention can further refine the sensitivity level. When the sensitivity level reaches a preset high sensitivity threshold, the input content is determined to have a sensitivity level exceeding level two, indicating that the input content contains high-risk sensitive information. Prioritizing the security of internal enterprise resources, the system no longer performs data security processing on the input, directly blocking and intercepting the input content, preventing it from being sent to the external large model. Simultaneously, the system will prompt the user that because the input content contains highly sensitive information, it cannot be sent to the external large model, and the user needs to re-edit the input content to ensure its internal compliance.

[0076] Step S304: Send the secure input content after data security processing to the external large model, and obtain the inference result generated by the external large model based on the secure input content.

[0077] Step S305: Identify the content to be determined in the reasoning result, wherein the content to be determined requires the input content to participate in the determination; based on the input content, obtain the content to be determined according to the reasoning logic corresponding to the content to be determined; assemble the obtained content to be determined and the determined content in the reasoning result to generate the response result.

[0078] For the inference results output by the external large model for secure input content after data security processing, in order to better match the user's original input content, the system's local secondary processing unit identifies the content to be determined that does not conform to the expected response result or cannot be directly used / called from the inference results based on the expected response result required by the original input content. For example, the user's expected response result contains a specific numerical value, while the inference result output by the external large model is a calculation method / formula; or, for example, the user's expected response result is a logical reasoning process, but because the sensitive information in the input content is generalized, the inference result output by the external large model is a logical reasoning framework. In these cases, the calculation method / formula and the logical reasoning framework are the content to be determined that the user cannot directly use / call.

[0079] Furthermore, the local secondary processing unit, according to the reasoning logic or determination method corresponding to the content to be determined, incorporates the original input content into the reasoning logic or determination method. For example, it substitutes sensitive information from the input content into the calculation method / formula for calculation to obtain the expected specific numerical response result; or, for example, it replaces the generalized content in the logical reasoning framework output by the external large model with sensitive information from the input content, and performs logical reasoning according to the logical reasoning framework to obtain a logical reasoning process that conforms to the expectations of the input content. Finally, the obtained determined content to be determined, along with other determined content in the reasoning result that does not require secondary processing, are assembled to obtain a response result that conforms to the user's original input content.

[0080] Understandably, this embodiment of the invention utilizes the reasoning capabilities of an external large model to provide secure input content, which has been processed to remove sensitive business information, to the external large model. This is equivalent to allowing the external large model to provide a problem-solving approach. The local secondary processing unit then uses sensitive real business data to perform secondary processing of business logic based on the problem-solving approach provided by the external large model. This ensures that, while maintaining internal data security, the final response result obtained by the user is an accurate result based on real business data, thus balancing the dual requirements of data security and leveraging the external large model.

[0081] Figure 4 This is a detailed flowchart illustrating the data processing method of this invention. A user activates an external large model within an internal system and enters input content in a dialog box. The system intercepts this input content. First, the intercepted input content is identified by its type, determining whether it is text, code, configuration, log, or a mixed type. Based on the identified content type, the corresponding coefficient Ct is determined. Next, the user's historical input is combined with the input content to form a combined input. Millisecond-level rule matching sensitivity detection is performed on the combined input content using regular expression matching rules, checksum generation rules, and trie rules. If sensitive information is detected, the sensitivity level of the input content is calculated based on the sensitivity Wi determined by the specific sensitive information, the coefficient Ct corresponding to the content type, and the coefficient Cu corresponding to the user's role characteristics. If no sensitive information is detected, semantic level detection (text semantic sensitivity detection and code semantic sensitivity detection) is performed on the input content in the combined input to identify sensitive information in the input content. Then, the sensitivity Wi of the input content is calculated based on the sensitive information. Combined with the coefficient Cu corresponding to the user's role characteristics and the coefficient Ct corresponding to the content type, (ΣWi) × Ct × Cu is calculated to obtain the sensitivity level of the input content.

[0082] If the sensitivity level is low-risk (Level 1), then a substitution or generalization approach is used to perform data security processing on the sensitive information, and the secure input content after data security processing is sent to the external large model. If the sensitivity level is medium-risk (Level 2), then a partial masking approach is used, and after user permission, the partially masked secure input content is sent to the external large model. If the sensitivity level is high-risk (Level 2 or higher), then the system no longer performs data security processing on the input content, directly blocks it, and does not send it to the external large model. Next, for the secure input content sent to the external large model, the local secondary processing unit generates a response result that matches the input content based on the inference result output by the external large model and the original input content.

[0083] Simultaneously, the system records the system processing event log corresponding to this user input content event and returns the log to the user to verify the correctness of the processing result. If the system detects updated feedback from the user regarding sensitive information, it updates the detection criteria and / or sensitivity level calculation parameters based on the updated feedback. Subsequent use of the updated detection criteria and / or calculation parameters improves the accuracy of the system's content sensitivity detection and sensitivity level calculation for future input content.

[0084] Figure 5 This is a schematic diagram of the main modules of a data processing apparatus according to an embodiment of the present invention. Figure 5 As shown, the data processing device 500 mainly includes a sensitivity detection module 501, a first processing module 502, and a second processing module 503.

[0085] Sensitivity detection module 501 is used to intercept the input content that the user is about to send to the external large model in response to the detection of the start of the external large model, perform content sensitivity detection on the input content, and identify sensitive information in the input content; The first processing module 502 is used to calculate the sensitivity level of the input content based on the sensitive information, and to perform data security processing on the sensitive information in the input content based on the sensitivity level. The second processing module 503 is used to send the secure input content after data security processing to the external large model, obtain the inference result generated by the external large model based on the secure input content, and generate the response result corresponding to the input content according to the inference result and the input content.

[0086] According to an embodiment of the present invention, the sensitivity detection module 501 is further configured to combine the historical inputs sent by the user to the external large model within a specified time with the input content to obtain a combined input; perform rule matching sensitivity detection on the combined input, and if no sensitive information is detected by the rule matching detection, perform semantic sensitivity detection on the combined input to identify explicit sensitive information in the input content; simulate the reasoning thought chain of the external large model on the combined input using a local detection model, and detect the sensitivity of the reasoning thought chain to identify implicit sensitive information in the input content; and determine the sensitive information in the input content based on the explicit and implicit sensitive information.

[0087] According to another embodiment of the present invention, the first processing module 502 is further configured to identify the content type of the input content, the content type including at least one of code type, configuration type, log type, and natural language type; calculate the sensitivity of the sensitive information based on the sensitive information, the frequency of occurrence of the sensitive information, and the timeliness of the sensitive information; and calculate the sensitivity level of the input content based on the sensitivity, the content type, and the user's role characteristics.

[0088] According to another embodiment of the present invention, the second processing module 503 is further configured to identify content to be determined in the reasoning result, wherein the content to be determined requires the input content to participate in the determination; based on the input content, obtain the content to be determined according to the reasoning logic corresponding to the content to be determined; assemble the obtained content to be determined and the determined content in the reasoning result to generate the response result.

[0089] According to another embodiment of the present invention, the first processing module 502 is further configured to replace or generalize the sensitive information when the sensitivity level is level one; and to partially cover the sensitive information or adjust the expression of the input content containing the sensitive information when the sensitivity level is level two, and send a corresponding processing permission request to the user to obtain the user's permission.

[0090] According to another embodiment of the present invention, the data processing device 500 further includes an input blocking processing module (not shown in the figure), which is used to intercept the sending action of the input content when the sensitivity level exceeds level two, and to prompt the user that there is highly sensitive information in the input content.

[0091] According to another embodiment of the present invention, the data processing device 500 further includes a feedback update module (not shown in the figure), which is used to update the detection criteria for content sensitivity detection and / or the calculation parameters for the sensitivity level based on the updated feedback information received regarding sensitive information.

[0092] Figure 6 This is an exemplary system architecture diagram in which embodiments of the present invention can be applied.

[0093] like Figure 6 As shown, system architecture 600 may include terminal devices 601, 602, and 603, a network 604, and a server 605. Network 604 serves as the medium for providing communication links between terminal devices 601, 602, and 603 and server 605. Network 604 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.

[0094] Users can use terminal devices 601, 602, and 603 to interact with server 605 via network 604 to receive or send messages, etc. Various communication client applications, such as data processing applications, can be installed on terminal devices 601, 602, and 603 (for example only).

[0095] Terminal devices 601, 602, and 603 can be various electronic devices with displays and web browsing capabilities, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0096] Server 605 can be a server providing various services, such as a backend management server supporting data processing performed by users using terminal devices 601, 602, and 603 (for example only). The backend management server can, in response to detecting the startup of an external large model, intercept the user's input content to be sent to the external large model, perform content sensitivity detection on the input content, identify sensitive information in the input content; calculate the sensitivity level of the input content based on the sensitive information, perform data security processing on the sensitive information in the input content based on the sensitivity level; send the secure input content after data security processing to the external large model, obtain the inference result generated by the external large model based on the secure input content, generate a response result corresponding to the input content based on the inference result and the input content, and feed back the response result to the terminal device.

[0097] It should be noted that the data processing method provided in the embodiments of the present invention is generally executed by the server 605, and correspondingly, the data processing device is generally located in the server 605.

[0098] It should be understood that Figure 6 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0099] The following is for reference. Figure 7 , Figure 7 This is a schematic diagram of the structure of a computer system suitable for implementing terminal devices or servers of the present invention. Figure 7 The terminal device or server shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of the present invention.

[0100] like Figure 7 As shown, the computer system 700 includes a central processing unit (CPU) 701, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 702 or programs loaded from storage section 708 into random access memory (RAM) 703. The RAM 703 also stores various programs and data required for the operation of the system 700. The CPU 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0101] The following components are connected to the I / O interface 705: an input section 706 including a keyboard, mouse, etc.; an output section 707 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a LAN card, modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the I / O interface 705 as needed. A removable medium 711, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 710 as needed so that computer programs read from it can be installed into the storage section 708 as needed.

[0102] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 709, and / or installed from removable medium 711. When the computer program is executed by central processing unit (CPU) 701, it performs the functions defined above in the system of this invention.

[0103] It should be noted that the computer-readable medium shown in this invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0104] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0105] The units described in the embodiments of the present invention can be implemented in software or in hardware. The described units can also be housed in a processor; for example, a processor can be described as including: a sensitivity detection module, a first processing module, and a second processing module.

[0106] In some cases, the names of these modules do not constitute a limitation on the module itself. For example, the sensitivity detection module can also be described as "a module used to intercept the user's input content to be sent to the external large model in response to the detection of the external large model's startup, perform content sensitivity detection on the input content, and identify sensitive information in the input content".

[0107] On the other hand, the present invention also provides a computer-readable medium, which may be included in the device described in the embodiments or may exist independently without being assembled into the device. The computer-readable medium carries one or more programs that, when executed by the device, cause the device to include: in response to detecting the startup of an external large model, intercepting user input content to be sent to the external large model; performing content sensitivity detection on the input content; identifying sensitive information in the input content; calculating the sensitivity level of the input content based on the sensitive information; performing data security processing on the sensitive information in the input content based on the sensitivity level; sending the data security-processed secure input content to the external large model; obtaining an inference result generated by the external large model based on the secure input content; and generating a response result corresponding to the input content based on the inference result and the input content.

[0108] The technical solution of the present invention has the following advantages or beneficial effects: by intercepting the user's input content to be sent to the external large model in response to the detection of the external large model's startup, performing content sensitivity detection on the input content, and identifying sensitive information in the input content; calculating the sensitivity level of the input content based on the sensitive information, and performing data security processing on the sensitive information in the input content based on the sensitivity level; sending the data security-processed secure input content to the external large model, obtaining the inference result generated by the external large model based on the secure input content, and generating the response result corresponding to the input content based on the inference result and the input content, the technical solution achieves accurate, fine-grained, and seamless data processing. Based on the sensitivity level, fine-grained data security processing is performed on the sensitive information in the input content, and the data security-processed secure input content is sent to the external large model. Finally, based on the inference result output by the external large model and the original input content, a response result that is imperceptible to the user is generated.

[0109] The specific embodiments described herein do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A data processing method, characterized in that, include: In response to the detection of the startup of an external large model, the system intercepts the user's input content to be sent to the external large model, performs content sensitivity detection on the input content, and identifies sensitive information in the input content. Based on the sensitive information, calculate the sensitivity level of the input content, and perform data security processing on the sensitive information in the input content according to the sensitivity level; The secure input content, after data security processing, is sent to the external large model. The inference result generated by the external large model based on the secure input content is obtained. Based on the inference result and the input content, a response result corresponding to the input content is generated.

2. The method according to claim 1, characterized in that, Perform content sensitivity detection on the input content to identify sensitive information in the input content, including: The user's historical inputs sent to the external large model within a specified time period are combined with the input content to obtain a combined input; The combined input is subjected to rule matching sensitivity detection, and if no sensitive information is detected by the rule matching detection, the combined input is subjected to semantic sensitivity detection to identify explicit sensitive information in the input content; The local detection model is used to simulate the reasoning thought chain of the external large model on the combined input, and the sensitivity of the reasoning thought chain is detected to identify implicit sensitive information in the input content. Based on the explicit and implicit sensitive information, the sensitive information in the input content is determined.

3. The method according to claim 1, characterized in that, Based on the sensitive information, the sensitivity level of the input content is calculated, including: Identify the content type of the input content, wherein the content type includes at least one of code type, configuration type, log type, and natural language type; Based on the sensitive information, the frequency of occurrence of the sensitive information, and the timeliness of the sensitive information, the sensitivity of the sensitive information is calculated. The sensitivity level of the input content is calculated based on the sensitivity, the content type, and the user's role characteristics.

4. The method according to claim 1, characterized in that, Based on the reasoning result and the input content, a response result corresponding to the input content is generated, including: Identify the content to be determined in the reasoning result, wherein the content to be determined requires the input content to participate in the determination; Based on the input content, the content to be determined is obtained according to the reasoning logic corresponding to the content to be determined; The obtained content to be determined and the determined content in the reasoning result are assembled to generate the response result.

5. The method according to claim 1, characterized in that, Based on the sensitivity level, data security processing is performed on the sensitive information in the input content, including: When the sensitivity level is Level 1, the sensitive information is replaced or generalized. When the sensitivity level is level two, the sensitive information is partially obscured, or the expression of the input content containing the sensitive information is adjusted, and a corresponding processing permission request is sent to the user to obtain the user's permission.

6. The method according to claim 5, characterized in that, The method further includes: If the sensitivity level exceeds level two, the sending of the input content is intercepted, and the user is notified that there is highly sensitive information in their input content.

7. The method according to claim 1, characterized in that, The method further includes: In response to receiving updated feedback information regarding sensitive information, the detection criteria for content sensitivity detection and / or the calculation parameters for the sensitivity level are updated based on the updated feedback information.

8. A data processing apparatus, characterized in that, include: The sensitivity detection module is used to intercept the user's input content to be sent to the external large model in response to the detection of the external large model's startup, perform content sensitivity detection on the input content, and identify sensitive information in the input content; The first processing module is used to calculate the sensitivity level of the input content based on the sensitive information, and to perform data security processing on the sensitive information in the input content based on the sensitivity level. The second processing module is used to send the secure input content after data security processing to the external large model, obtain the inference result generated by the external large model based on the secure input content, and generate the response result corresponding to the input content according to the inference result and the input content.

9. A mobile electronic device terminal, characterized in that, include: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-7.

10. A computer-readable medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-7.

11. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-7.

Images