Method and device for desensitizing intellectual property in engineering data assets
By building a sensitive feature library and using GRU neural network and generative artificial intelligence models to automatically identify and desensitize enterprise-level intellectual property rights, the problem of intellectual property desensitization in the existing technology is solved, and efficient and accurate intellectual property protection is achieved.
Patent Information
- Application Number
- CN202411588215.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-08
- Publication Date
- 2025-02-28
AI Technical Summary
The existing technology has problems such as relying on manual operations, inefficiency and poor desensitization in the desensitization process of enterprise-level intellectual property, and lacks systematic solutions.
By collecting intellectual property materials in the enterprise's engineering data assets, building a sensitive feature library, and using GRU neural network models and generative artificial intelligence models, we automatically identify and desensitize sensitive information in pending data assets.
It realizes accurate and efficient intellectual property desensitization of enterprise-level data assets, improves the accuracy and efficiency of desensitization, effectively protects enterprise-level non-public intellectual property rights, and reduces the risk of information leakage.
Smart Images
Figure CN119538308A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a method and device for desensitizing intellectual property rights in engineering data assets. Background Art
[0002] Intellectual property protection in enterprise data assets is increasingly valued. In the field of engineering, enterprises have a large amount of non-public intellectual property materials, which are an important manifestation of the core competitiveness of enterprises. Once leaked, it will cause immeasurable losses to the enterprise. Therefore, how to effectively protect these intellectual property rights and prevent the leakage of key information has become the focus of enterprises.
[0003] Under existing technical conditions, there are some desensitization methods for text and images, but there are often problems such as the desensitization process relying on manual operation, poor desensitization effect, and low efficiency. In addition, there is a lack of systematic desensitization solutions for enterprise-level intellectual property. Summary of the invention
[0004] The purpose of the present invention is to provide a method and device for desensitizing intellectual property rights in engineering data assets.
[0005] In order to solve the above problems, the present invention provides a method for desensitizing intellectual property rights in engineering data assets, comprising: Step 1: Collect intellectual property materials in the enterprise's engineering data assets, and build a sensitive feature library based on the intellectual property materials; Step 2: Based on the sensitive feature library, train an intelligent recognition model to construct a feature vector of key information; Step 3: Based on the feature vector of the key information and the trained GRU neural network model M, a secondary positioning method is used to find the sensitive information in the data assets to be desensitized; Step 4: Based on the sensitive feature library, a generative artificial intelligence model is trained to desensitize the sensitive information in the data assets to be desensitized.
[0006] Furthermore, in the above method, step 1: collecting intellectual property materials in the engineering data assets of the enterprise, and building a sensitive feature library based on the intellectual property materials, including: Step 1.1: Collect non-public intellectual property materials from enterprise-level engineering data assets; Step 1.2: Use a large language model to learn the intellectual property material and extract key information K i , where i represents the serial number of key information in a certain intellectual property type in the sensitive intellectual property database; Step 1.3: Extract the key information K i , classified by intellectual property type, forming a table containing <intellectual property type-key information K iCollection> of sensitive feature libraries.
[0007] Furthermore, in the above method, the key information K i , including: key innovation points, key processes and key confidential content.
[0008] Furthermore, in the above method, step 2: based on the sensitive feature library, training an intelligent recognition model to construct a feature vector of key information includes: Step 2.1: Using the sensitive feature library, train the GRU neural network model M so that the GRU neural network model M meets the preset requirements to obtain a trained GRU neural network model M; Step 2.2: Describe the key information K in the sensitive feature library i Run the trained GRU neural network model M on the machine, for each key information K i , retain the calculation result vector V of the penultimate layer of the M neural network i , which is the key information K i The corresponding feature vector.
[0009] Furthermore, in the above method, making the GRU neural network model M meet the preset requirements includes: The accuracy, recall rate and detection precision of the GRU neural network model M training all meet the preset requirements.
[0010] Furthermore, in the above method, step 3: based on the feature vector of the key information and the trained GRU neural network model M, and using a secondary positioning method, finding the sensitive information in the data asset to be desensitized includes: Step 3.1: Divide the text data assets into paragraphs, and then run the trained GRU neural network model M on each of the divided paragraphs. If the neural network model M detects that the paragraph belongs to any type of intellectual property, the paragraph is located as a sensitive information candidate paragraph P; Step 3.2: For the sensitive information candidate paragraph P, retain the calculation result vector V of the penultimate layer of the neural network model M P , the vector V P Each feature vector V of all sensitive intellectual property libraries i Compare and calculate the Pearson correlation coefficient. If V P and any V i If the correlation coefficient is greater than a preset first threshold, the paragraph is located as the final sensitive information location.
[0011] Furthermore, in the above method, step 4: training a generative artificial intelligence model based on the sensitive feature library to desensitize sensitive information in the data assets to be desensitized includes: Step 4.1: Using the sensitive feature library, training sensitive features to generate an artificial intelligence model G, wherein the neural network model M is used as a discriminator, and another GRU network with the same structure as the neural network model M is used as a generator; training until the discriminator's judgment accuracy rate for the generated text is lower than a preset third threshold; Step 4.2: For the sensitive information in the located sensitive information location, use the trained generative artificial intelligence model G to generate a realistic non-sensitive description text about the intellectual property rights, and use the realistic non-sensitive description text about the intellectual property rights to replace the sensitive information in the sensitive information location.
[0012] Furthermore, in the above method, the first threshold is 0.8, and the third threshold is 50%. According to another aspect of the present invention, there is also provided a computer-readable storage medium having computer-executable instructions stored thereon, wherein when the computer-executable instructions are executed by a processor, the processor is caused to: execute the method described in any of the above embodiments. According to another aspect of the present invention, there is also provided a computer device, comprising: Processor; and A memory arranged to store computer executable instructions, which, when executed, cause the processor to: execute the method described in any one of the above embodiments. Compared with the prior art, the present invention collects intellectual property materials in the engineering data assets of the enterprise, and builds a sensitive feature library based on the intellectual property materials; based on the sensitive feature library, trains an intelligent recognition model to construct a feature vector of key information; based on the feature vector of the key information and the trained GRU neural network model M, and adopts a secondary positioning method to find the sensitive information in the data assets to be desensitized; based on the sensitive feature library, trains a generative artificial intelligence model to desensitize the sensitive information in the data assets to be desensitized. The present invention adopts a large language model and generative technology, which can perform intellectual property desensitization of images and texts, realize accurate and efficient intellectual property desensitization of enterprise-level data assets, improve the accuracy and efficiency of intellectual property desensitization, can effectively protect enterprise-level non-public intellectual property, reduce the risk of information leakage, and can provide an efficient, intelligent and comprehensive solution for enterprise-level intellectual property protection. BRIEF DESCRIPTION OF THE DRAWINGS
[0013] Figure 1 It is a flow chart of a method for desensitizing intellectual property rights in engineering data assets according to an embodiment of the present invention. DETAILED DESCRIPTION
[0014] The present invention is further described in detail below in conjunction with the accompanying drawings.
[0015] In a typical configuration of the present application, the terminal, the device of the service network and the trusted party all include one or more processors (CPU), input / output interfaces, network interfaces and memories.
[0016] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.
[0017] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include non-transitory media such as modulated data signals and carrier waves.
[0018] like Figure 1 As shown, the present invention provides a method for desensitizing intellectual property rights in engineering data assets, the method comprising: step 1: collecting intellectual property materials in the engineering data assets of an enterprise, and building a sensitive feature library based on the intellectual property materials: step 1.1: non-public intellectual property materials in enterprise-level engineering data assets can be collected and classified into enterprise method libraries, special construction plans, patent applications, award materials and technical monographs, etc.; Step 1.2: Use a large language model to learn the intellectual property material and extract key information K i , where i represents the serial number of key information of a certain intellectual property type in the sensitive intellectual property database; preferably, the key information K i , including: key innovation points, key processes and key confidential contents; Step 1.3: Extract the key information K i , classified by intellectual property type, forming a table containing <intellectual property type-key information K i Collection> of sensitive feature libraries.
[0019] For example, a large language model can be used to extract key text from 100 documents on internal work methods of an enterprise to form a set of 100 key information of the "sensitive work method" intellectual property type as part of the sensitive feature library.
[0020] Step 2: Based on the sensitive feature library, train the intelligent recognition model and construct the feature vector of key information: Step 2.1: Use the sensitive feature library to train the GRU neural network model M, so that the accuracy, recall rate and detection accuracy of the GRU neural network model M training meet the preset requirements to obtain the trained GRU neural network model M; Step 2.2: In the key information K of the sensitive feature library i Run the trained GRU neural network model M on the machine, for each key information K i , retain the calculation result vector V of the penultimate layer of the M neural network i , which is the key information K i The corresponding feature vector.
[0021] Specifically, the above-mentioned <Intellectual Property Type-Key Information K i The sensitive feature library of the collection > is used to train the GRU neural network model M. The output of the model in the second to last step is a vector, and the last step is to determine the type. In general, only the final result is needed, but the technical solution here also stores the intermediate result of the second to last step as the most representative feature vector for secondary comparison.
[0022] Step 3: Based on the feature vector of the key information and the trained GRU neural network model M, a secondary positioning method is used to find the sensitive information in the data assets to be desensitized; Step 3.1: Divide the text data assets into paragraphs, and then run the trained GRU neural network model M on each of the divided paragraphs. If the neural network model M detects that the paragraph belongs to any type of intellectual property, the paragraph is located as a sensitive information candidate paragraph P; Step 3.2: For the sensitive information candidate paragraph P, retain the calculation result vector V of the penultimate layer of the neural network model M P , the vector V P Each feature vector V of all sensitive intellectual property libraries i Compare and calculate the Pearson correlation coefficient. If V P and any V i If the correlation coefficient is greater than a preset first threshold value such as 0.8, the paragraph is located as the final sensitive information location.
[0023] For example, in step 3.1, a paragraph is input into the paragraph M, and the probability that it belongs to a certain "intellectual property type" is greater than a second preset threshold value such as 90%, then the paragraph is located as a sensitive information candidate paragraph P; Therefore, in step 3.2, the paragraph is judged to be a sensitive information candidate paragraph P, and the second-to-last intermediate result of the paragraph M is taken out, which is V P , V P Compare with all feature vectors in the library and calculate the correlation coefficient. Only those greater than 0.8 are truly sensitive information. Otherwise, desensitization is not required.
[0024] Step 4: Based on the sensitive feature library, a generative artificial intelligence model is trained to desensitize sensitive information in the data assets to be desensitized: Step 4.1: Use the sensitive feature library to train sensitive features to generate an artificial intelligence model G, wherein the neural network model M is used as a discriminator and another GRU network with the same structure as the neural network model M is used as a generator; train until the discriminator's judgment accuracy on the generated text is lower than a preset third threshold value such as 50%.
[0025] Step 4.2: For the sensitive information in the located sensitive information position, use the trained generative artificial intelligence model G to generate a realistic non-sensitive description text about intellectual property rights, and use the realistic non-sensitive description text about intellectual property rights to replace the sensitive information in the sensitive information position, ultimately achieving the desensitization effect.
[0026] For example, the generative artificial intelligence model G can be trained to replace the original sensitive information by generating a realistic paragraph that looks similar but is actually quite random. According to another aspect of the present invention, there is also provided a computer-readable storage medium having computer-executable instructions stored thereon, wherein when the computer-executable instructions are executed by a processor, the processor is caused to: execute the method described in any of the above embodiments. According to another aspect of the present invention, there is also provided a computer device, comprising: Processor; and A memory arranged to store computer executable instructions, which, when executed, cause the processor to: execute the method described in any one of the above embodiments. In summary, the present invention adopts a large language model and generative technology to perform intellectual property desensitization of images and texts, realize accurate and efficient intellectual property desensitization of enterprise-level data assets, improve the accuracy and efficiency of intellectual property desensitization, effectively protect enterprise-level non-public intellectual property rights, reduce the risk of information leakage, and provide an efficient, intelligent and comprehensive solution for enterprise-level intellectual property protection.
[0027] The detailed contents of the device embodiments of the present invention can be found in the corresponding parts of the method embodiments, which will not be described in detail here.
[0028] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.
[0029] It should be noted that the present invention can be implemented in software and / or a combination of software and hardware, for example, can be implemented using an application specific integrated circuit (ASIC), a general purpose computer or any other similar hardware device. In one embodiment, the software program of the present invention can be executed by a processor to implement the steps or functions described above. Similarly, the software program of the present invention (including related data structures) can be stored in a computer readable recording medium, for example, a RAM memory, a magnetic or optical drive or a floppy disk and the like. In addition, some steps or functions of the present invention can be implemented using hardware, for example, as a circuit that cooperates with a processor to perform each step or function.
[0030] In addition, a part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. The program instruction for calling the method of the present invention may be stored in a fixed or removable recording medium, and / or transmitted through a data stream in a broadcast or other signal-bearing medium, and / or stored in a working memory of a computer device that runs according to the program instruction. Here, according to an embodiment of the present invention, a device is included, the device including a memory for storing computer program instructions and a processor for executing program instructions, wherein, when the computer program instruction is executed by the processor, the device is triggered to run the method and / or technical solution based on the aforementioned multiple embodiments of the present invention.
[0031] It is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-restrictive, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any figure mark in the claims should not be regarded as limiting the claims involved. In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the device claim can also be implemented by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any particular order.
Claims
1. A method for desensitizing intellectual property rights in engineering data assets, characterized in that: include: Step 1: Collect intellectual property materials in the enterprise's engineering data assets, and build a sensitive feature library based on the intellectual property materials; Step 2: Based on the sensitive feature library, train an intelligent recognition model to construct a feature vector of key information; Step 3: Based on the feature vector of the key information and the trained GRU neural network model M, a secondary positioning method is used to find the sensitive information in the data assets to be desensitized; Step 4: Based on the sensitive feature library, a generative artificial intelligence model is trained to desensitize the sensitive information in the data assets to be desensitized.
2. The method for desensitizing intellectual property rights in engineering data assets according to claim 1, characterized in that: Step 1: Collect the intellectual property materials in the enterprise's engineering data assets and build a sensitive feature library based on the intellectual property materials, including: Step 1.1: Collect non-public intellectual property materials from enterprise-level engineering data assets; Step 1.2: Use a large language model to learn the intellectual property material and extract key information K i , where i represents the serial number of key information in a certain intellectual property type in the sensitive intellectual property database; Step 1.3: Extract the key information K i , classified by intellectual property type, forming a table containing <intellectual property type-key information K i Collection> of sensitive feature libraries.
3. The method for desensitizing intellectual property rights in engineering data assets according to claim 2, characterized in that: The key information K i , including: key innovation points, key processes and key confidential content.
4. The method for desensitizing intellectual property rights in engineering data assets according to claim 2, characterized in that: Step 2: Based on the sensitive feature library, train an intelligent recognition model to construct a feature vector of key information, including: Step 2.1: Using the sensitive feature library, train the GRU neural network model M so that the GRU neural network model M meets the preset requirements to obtain a trained GRU neural network model M; Step 2.2: Describe the key information K in the sensitive feature library i Run the trained GRU neural network model M on the machine, for each key information K i , retain the calculation result vector V of the penultimate layer of the M neural network i , which is the key information K i The corresponding feature vector.
5. The method for desensitizing intellectual property rights in engineering data assets according to claim 4, characterized in that: Make the GRU neural network model M meet the preset requirements, including: The accuracy, recall rate and detection precision of the GRU neural network model M training all meet the preset requirements.
6. The method for desensitizing intellectual property rights in engineering data assets according to claim 4, characterized in that: Step 3: Based on the feature vector of the key information and the trained GRU neural network model M, a secondary positioning method is used to find the sensitive information in the data assets to be desensitized, including: Step 3.1: Divide the text data assets into paragraphs, and then run the trained GRU neural network model M on each of the divided paragraphs. If the neural network model M detects that the paragraph belongs to any type of intellectual property, the paragraph is located as a sensitive information candidate paragraph P; Step 3.2: For the sensitive information candidate paragraph P, retain the calculation result vector V of the penultimate layer of the neural network model M P , the vector V P Each feature vector V of all sensitive intellectual property libraries i Compare and calculate the Pearson correlation coefficient. If V P and any V i If the correlation coefficient is greater than a preset first threshold, the paragraph is located as the final sensitive information location.
7. The method for desensitizing intellectual property rights in engineering data assets according to claim 6, characterized in that: Step 4: Based on the sensitive feature library, a generative artificial intelligence model is trained to desensitize sensitive information in the data assets to be desensitized, including: Step 4.1: Using the sensitive feature library, sensitive features are trained to generate an artificial intelligence model G, wherein the neural network model M is used as a discriminator, and another GRU network with the same structure as the neural network model M is used as a generator; the training is performed until the discriminator's judgment accuracy rate on the generated text is lower than a preset third threshold; Step 4.2: For the sensitive information in the located sensitive information location, use the trained generative artificial intelligence model G to generate a realistic non-sensitive description text about the intellectual property rights, and use the realistic non-sensitive description text about the intellectual property rights to replace the sensitive information in the sensitive information location.
8. The method for desensitizing intellectual property rights in engineering data assets according to claim 1, characterized in that: The first threshold is 0.8, and the third threshold is 50%.
9. A computer-readable storage medium having computer-executable instructions stored thereon, wherein: When the computer executable instructions are executed by a processor, the processor is caused to: perform the method according to any one of claims 1 to 8.
10. A computer device, wherein: include: processor; as well as A memory arranged to store computer executable instructions which, when executed, cause the processor to: perform a method as claimed in any one of claims 1 to 8.