Audit tag configuration strategy generation method, device, equipment and storage medium

Abstract

The present invention provides a method, device, equipment and storage medium for generating an audit label configuration strategy, including: obtaining an interception standard document and an audit standard specification document; extracting multiple text information and image information based on the audit standard specification document and the interception standard document; generating query instruction information based on each text information and each image information; inputting the query instruction information into a preset question and answer model to obtain an audit label configuration strategy. The present invention extracts text information and image information based on the interception standard document and the audit standard specification document, and then generates query instruction information by combining the text information and the image information, and inputs the query instruction information into an open source preset question and answer model to obtain an audit label configuration strategy, thereby eliminating the need to manually configure audit labels according to standards, reducing manpower and time costs, and the open source preset question and answer model has strong understanding and reasoning capabilities, making the generated audit label configuration strategy more comprehensive.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method, device, equipment and storage medium for generating an audit tag configuration strategy. Background Art

[0002] Currently, there are sensitive contents such as violence and vulgarity in various types of information on the Internet. In order to purify the network environment and create a clear network space, content risk control review is very important. In the past, content risk control review relied purely on manual review, which was time-consuming and laborious.

[0003] In recent years, with the development of machine learning and deep learning algorithms, machine review has become the main force of content risk control review due to its good cost-effectiveness. However, the current machine review system for text and image content risk control still needs to rely on a lot of manual work, especially for B-side customers. In the face of different customers and different scenarios, it is necessary to arrange experienced configuration personnel to fully communicate with customers and manually configure different review labels to obtain good interception effects. However, manually configuring review labels is very costly in terms of manpower and time, and requires configuration personnel to accurately and deeply understand the capabilities of each review label. Once there is a deviation in understanding, it will affect the final review effect. Summary of the invention

[0004] The present invention provides a method, device, equipment and storage medium for generating an audit label configuration strategy, aiming to solve the technical problem that manual configuration of audit labels has high manpower and time costs, and requires configuration personnel to have an accurate and in-depth understanding of the capabilities of each audit label. Once there is a deviation in understanding, it will affect the final audit effect.

[0005] The present invention provides a method for generating an audit tag configuration strategy, comprising:

[0006] Obtain the interception standard documents of target customers and the audit standard specification documents of products;

[0007] Based on the audit standard specification document and the interception standard document, extract multiple text information and multiple image information;

[0008] Generate inquiry instruction information based on each of the text information and each of the picture information;

[0009] The query instruction information is input into a preset question-answering model to obtain an audit label configuration strategy output by the preset question-answering model.

[0010] According to a method for generating an audit label configuration strategy provided by the present invention, the method extracts a plurality of text information and a plurality of image information based on the audit standard specification document and the interception standard document, including:

[0011] Performing text extraction and image extraction on the audit standard specification document and the interception standard document in sequence to obtain a plurality of text extraction information and image information associated with the text extraction information;

[0012] Semantic segmentation is performed on each of the text extraction information to obtain each of the text information.

[0013] According to a method for generating an audit label configuration strategy provided by the present invention, the query instruction information is generated based on each of the text information and each of the picture information, including:

[0014] Encoding each of the picture information to obtain picture encoding information;

[0015] Based on the picture coding information and the text information associated therewith, query instruction information in a preset query format is generated.

[0016] According to a method for generating an audit label configuration strategy provided by the present invention, the interception standard document and the audit standard specification document include one or more of multi-level label information, definition description information and sample information.

[0017] According to a method for generating an audit tag configuration strategy provided by the present invention, after inputting the query instruction information into a preset question-answer model to obtain the audit tag configuration strategy output by the preset question-answer model, the method further includes:

[0018] Acquire training samples of different sensitive content types, wherein the training samples are configured with sample labels corresponding to the review label configuration strategy;

[0019] Iteratively training a preset content review model based on each of the training samples and their corresponding sample labels;

[0020] Use the trained content review model to review the content to be reviewed and obtain the content review results.

[0021] According to a method for generating an audit tag configuration strategy provided by the present invention, before inputting the query instruction information into a preset question-answer model to obtain the audit tag configuration strategy output by the preset question-answer model, the method further includes:

[0022] Get several different audit topics;

[0023] For any of the audit questions: repeatedly input the audit question into the preset question-answering model to obtain a plurality of first feedback information output by the preset question-answering model,

[0024] Sorting the first feedback information to train a reward model according to the sorting result;

[0025] For any of the audit questions: inputting the audit question into the preset question-answering model to obtain second feedback information output by the preset question-answering model;

[0026] Using the trained reward model to score the second feedback information;

[0027] Based on the scoring results and the review questions, the parameters of the preset question and answer model are adjusted.

[0028] According to a method for generating an audit tag configuration strategy provided by the present invention, before inputting the query instruction information into a preset question-answer model to obtain the audit tag configuration strategy output by the preset question-answer model, the method further includes:

[0029] Obtaining a plurality of inquiry questions with different interception standards and audits, wherein each of the inquiry questions is configured with answer information;

[0030] Based on each of the inquiry questions and the corresponding answer information, parameters of the preset question-answering model are adjusted.

[0031] The present invention also provides an audit tag configuration strategy generation device, comprising:

[0032] The acquisition module is used to obtain the interception standard documents of target customers and the audit standard specification documents of products;

[0033] An extraction module, used for extracting a plurality of text information and a plurality of image information based on the audit standard specification document and the interception standard document;

[0034] A generating module, used for generating inquiry instruction information based on each of the text information and each of the picture information;

[0035] The label configuration module is used to input the query instruction information into a preset question-answering model to obtain the audit label configuration strategy output by the preset question-answering model.

[0036] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for generating an audit tag configuration strategy as described above is implemented.

[0037] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the audit tag configuration strategy generation method as described in any one of the above is implemented.

[0038] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the audit tag configuration strategy generation method described in any one of the above is implemented.

[0039] The audit label configuration strategy generation method, device, equipment and storage medium provided by the present invention include: obtaining the interception standard document of the target customer and the audit standard specification document of the product; extracting multiple text information and multiple image information based on the audit standard specification document and the interception standard document; generating query instruction information based on each of the text information and each of the image information; inputting the query instruction information into a preset question-and-answer model to obtain the audit label configuration strategy output by the preset question-and-answer model. The present invention extracts text and images from the interception standard document and the audit standard specification document of the product to obtain multiple text information and multiple image information, and then generates query instruction information by combining each of the text information and each of the image information, thereby inputting the query instruction information into an open source preset question-and-answer model to obtain the audit label configuration strategy, thereby eliminating the need to manually configure the audit label configuration strategy according to the standard, reducing manpower and time costs, and the understanding ability and reasoning ability of the open source preset question-and-answer model are relatively stable, making the generated audit label configuration strategy more comprehensive, thereby reducing the situation where the manual configuration of the audit label configuration strategy has deviations that affect the audit effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced one by one below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 It is a flowchart of the audit tag configuration strategy generation method provided by the present invention;

[0042] Figure 2 It is a complete schematic diagram of the audit tag configuration strategy generation method provided by the present invention;

[0043] Figure 3 It is a structural schematic diagram of the audit tag configuration strategy generation device provided by the present invention;

[0044] Figure 4 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] The terms used in one or more embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit one or more embodiments of the present invention. The singular forms of "a", "said" and "the" used in one or more embodiments of the present invention are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in one or more embodiments of the present invention refers to and includes any or all possible combinations of one or more associated listed items.

[0047] It should be understood that, although the terms first, second, etc. may be used to describe various information in one or more embodiments of the present invention, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of the present invention, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when the time of".

[0048] Currently, there are sensitive contents such as violence and vulgarity in various types of information on the Internet. In order to purify the network environment and create a clear network space, content risk control review is very important. In the past, content risk control review relied purely on manual review, which was time-consuming and laborious.

[0049] In recent years, with the development of machine learning and deep learning algorithms, machine review has become the main force of content risk control review with good cost performance. However, the current machine review system for graphic content risk control still needs to rely on a lot of manual work, especially for B-side customers. In the face of different customers and different scenarios, it is necessary to arrange experienced configuration personnel to fully communicate with customers and manually configure different review label configuration strategies to obtain good interception effects. The cost in manpower and time is very high, and it requires the configuration personnel to have an accurate and in-depth understanding of each review label configuration strategy. Once there is a deviation in understanding, it will affect the final review effect.

[0050] In view of the above problems, the present invention provides the following embodiments. Specifically, Figure 1 FIG. 1 is a flow chart of the method for generating the audit tag configuration strategy provided by the present invention. Figure 1 As shown, the audit tag configuration strategy generation method includes:

[0051] Step S11, obtaining the interception standard document of the target customer and the audit standard specification document of the product;

[0052] It should be noted that the customer's interception standard document focuses on the customer's needs, expectations and standards. This may include customer specifications, requirements and standardized processes to ensure that the product or service meets their expectations. The customer's interception standard document focuses on customer and market needs and may involve stakeholders related to customers, sales and marketing.

[0053] It should be noted that the product audit standard specification document focuses more on the product's specifications, quality standards, manufacturing processes, and compliance with specific industry or regulatory requirements. The product audit standard specification document may be developed by internal departments such as product development, quality control, or manufacturing departments.

[0054] It should be noted that customers can fill in the document in the specified format, and then the policy configurator will check and organize it into a certain form of interception standard document and audit standard specification document. Optionally, the interception standard document and audit standard specification document include one or more of multi-level label information, definition description information and sample information. The multi-level label information represents the type label of the sensitive content, the definition description information represents the detailed description information of the type of sensitive content, and the sample information represents a detailed example or sample diagram. Optionally, the document format can refer to Table 1.

[0055] Table 1

[0056]

[0057] Step S12, extracting a plurality of text information and a plurality of image information based on the audit standard specification document and the interception standard document;

[0058] Specifically, since the interception standard document and the product audit standard specification document are documents integrated in a certain form, in this embodiment, the audit standard specification document and the interception standard document can be respectively subjected to text and image recognition in sequence to obtain a number of text extraction information and image information associated with the text extraction information. For example, in the order of primary tags, secondary tags, definition description information and sample information, the following are identified: advertisements, competitors, competitors' products, trademarks, logos and other contents, text extraction information of product A and corresponding image information of trademark and logo sample images. And semantic segmentation is performed on each of the text extraction information to obtain each of the text information.

[0059] Step S13, generating inquiry instruction information based on each of the text information and each of the picture information;

[0060] Specifically, each of the picture information is encoded to obtain picture encoding information, for example, the picture information is encoded through a layer of linear change layer. Then, based on the picture encoding information and its associated text information, a query instruction information of a preset query format is generated. Optionally, the preset query format can be set according to actual conditions, for example: the review standards are now defined as follows: Tags: picture sexy - revealing clothing, defined as female breasts exposed; picture advertisement - mobile phone logo, defined as the logo of the mobile phone brand.

[0061] The query command information that can be generated is as follows: The current customer's image blocking standards are: 1. Blocking of women's breasts exposed, such as: [0.123, 0.862, 0.411...]; 2. Blocking of all competing content involving brand A mobile phones, including products, trademarks, and logos. Please output the recommended audit label configuration strategy. If there is no corresponding label capability, please specify it clearly.

[0062] Step S14: input the query instruction information into a preset question-answering model to obtain an audit tag configuration strategy output by the preset question-answering model.

[0063] It should be noted that the preset question-and-answer model is an open source large model, such as ChatGPT and BERT models. Existing open source large models can understand human language very well. Therefore, by inputting the query instruction information into the preset question-and-answer model, the preset question-and-answer model can output the audit label configuration strategy corresponding to the target customer.

[0064] The embodiment of the present invention, through the above scheme, includes: obtaining the interception standard document of the target customer and the audit standard specification document of the product; extracting multiple text information and multiple image information based on the audit standard specification document and the interception standard document; generating query instruction information based on each of the text information and each of the image information; inputting the query instruction information into the preset question-answer model to obtain the audit label configuration strategy output by the preset question-answer model. By extracting text and images from the interception standard document and the audit standard specification document of the product, multiple text information and multiple image information are obtained, and then combining each of the text information and each of the image information to generate query instruction information, so that the query instruction information is input into the open source preset question-answer model to obtain the audit label configuration strategy, thereby eliminating the need to manually configure the audit label configuration strategy according to the standard, reducing manpower and time costs, and the understanding ability and reasoning ability of the open source preset question-answer model are relatively stable, making the generated audit label configuration strategy more comprehensive, thereby reducing the situation where there is deviation in the manual configuration of the audit label that affects the audit effect.

[0065] In one embodiment of the present invention, the generating of the query instruction information based on each of the text information and each of the picture information includes:

[0066] Each of the picture information is encoded to obtain picture encoding information; and based on the picture encoding information and its associated text information, query instruction information in a preset query format is generated.

[0067] Specifically, each of the picture information is encoded to obtain picture encoding information, and then the picture encoding information is weighted, and then the weighted picture encoding information and its associated text information are combined to generate query instruction information in a preset query format.

[0068] In one embodiment of the present invention, before inputting the query instruction information into a preset question-answer model to obtain the audit tag configuration strategy output by the preset question-answer model, the method further includes:

[0069] Obtain a number of different review questions; for any of the review questions: repeatedly input the review question into the preset question and answer model to obtain a number of first feedback information output by the preset question and answer model, and sort the first feedback information to train the reward model according to the sorting result; for any of the review questions: input the review question into the preset question and answer model to obtain second feedback information output by the preset question and answer model; score the second feedback information using the trained reward model; and adjust the parameters of the preset question and answer model based on the scoring result and the review question.

[0070] It should be noted that in actual use, the interception standards for various customers and scenarios are relatively complex. Considering that the open source big model is not specifically aimed at content review scenarios, it is necessary to fine-tune the open source big model.

[0071] Specifically, first obtain a number of different audit questions, and then perform the following steps for any of the audit questions: repeatedly input the audit question into the preset question-and-answer model to obtain a number of first feedback information output by the preset question-and-answer model, and then obtain the sorting result of manually sorting the first feedback information corresponding to any of the audit questions. Further, use the sorting result and the corresponding reward information to train the reward model. This usually involves supervised learning or reinforcement learning methods. In reinforcement learning, an algorithm similar to policy gradient may be used to update the model parameters by maximizing the expected reward to obtain a trained reward model. Further: for any of the audit questions: input the audit question into the preset question-and-answer model to obtain the second feedback information output by the preset question-and-answer model; use the trained reward model to score the second feedback information. Then use the scoring result of the reward model to adjust the parameters of the preset question-and-answer model. This can be done by updating the parameters of the model according to the direction of the reward through optimization algorithms such as gradient descent.

[0072] The embodiment of the present invention adjusts the parameters of the preset question and answer model so that the preset question and answer model can learn more information corresponding to the content review scenario, thereby improving the accuracy of the model's response.

[0073] In one embodiment of the present invention, before inputting the query instruction information into a preset question-answer model to obtain the audit tag configuration strategy output by the preset question-answer model, the method further includes:

[0074] A plurality of inquiry questions with different interception standards and audits are obtained, wherein each of the inquiry questions is configured with answer information; and based on each of the inquiry questions and the corresponding answer information, the parameters of the preset question-answer model are adjusted.

[0075] Specifically, a number of inquiry questions with different interception standards and audits are obtained, wherein each of the inquiry questions is configured with manually set answer information, and then based on each of the inquiry questions and the corresponding answer information, the parameters of the preset question and answer model are adjusted.

[0076] The embodiment of the present invention adjusts the parameters of the preset question and answer model so that the preset question and answer model can learn more information corresponding to the content review scenario, thereby improving the accuracy of the model's response.

[0077] In one embodiment of the present invention, after inputting the query instruction information into a preset question-answer model to obtain the audit tag configuration strategy output by the preset question-answer model, the method further includes:

[0078] Obtain training samples of different sensitive content types, wherein the training samples are configured with sample labels corresponding to the review label configuration strategy; iteratively train a preset content review model based on each of the training samples and their corresponding sample labels; and use the trained content review model to review the content to be reviewed to obtain a content review result.

[0079] Specifically, first, a large number of training samples containing different types of sensitive content are obtained. These training samples need to be labeled, that is, a corresponding audit label configuration strategy (that is, the sample label in this embodiment) is assigned to each training sample to indicate the type of sensitive content present in the training sample. Then, based on each of the training samples and their corresponding sample labels, a preset content audit model is iteratively trained. Thus, in the content audit process, the content to be audited is obtained, and then the trained content audit model is used to audit the content to be audited to obtain the content audit result.

[0080] Reference Figure 2 , Figure 2It is a complete schematic diagram of the audit label configuration strategy generation method provided by the present invention. Specifically, firstly, the interception standard document of the customer and the audit standard specification document of the product integrated in a certain form are obtained, and then multiple text information and multiple image information are sequentially extracted from the audit standard specification document and the interception standard document, and further, based on each of the text information and each of the image information, query instruction information is generated, so that the query instruction information is input into the preset question and answer model to obtain the audit label configuration strategy output by the preset question and answer model.

[0081] The audit tag configuration policy generating device provided by the present invention is described below. The audit tag configuration policy generating device described below and the audit tag configuration policy generating method described above can be referred to each other.

[0082] Figure 3 is a schematic diagram of the structure of the audit tag configuration strategy generation device provided by the present invention, such as Figure 3 As shown, an audit tag configuration strategy generation device according to an embodiment of the present invention includes:

[0083] An acquisition module 21 is used to acquire the interception standard document of the target customer and the audit standard specification document of the product;

[0084] An extraction module 22, configured to extract a plurality of text information and a plurality of image information based on the audit standard specification document and the interception standard document;

[0085] A generating module 23, used for generating inquiry instruction information based on each of the text information and each of the picture information;

[0086] The label configuration module is used to input the query instruction information into a preset question-answering model to obtain the audit label configuration strategy output by the preset question-answering model.

[0087] The audit tag configuration strategy generating device further includes:

[0088] Performing text extraction and image extraction on the audit standard specification document and the interception standard document in sequence to obtain a plurality of text extraction information and image information associated with the text extraction information;

[0089] Semantic segmentation is performed on each of the text extraction information to obtain each of the text information.

[0090] The audit tag configuration strategy generating device further includes:

[0091] Encoding each of the picture information to obtain picture encoding information;

[0092] Based on the picture coding information and the text information associated therewith, query instruction information in a preset query format is generated.

[0093] The audit tag configuration strategy generating device further includes:

[0094] The interception standard document and the audit standard specification document include one or more of multi-level label information, definition description information and sample information.

[0095] The audit tag configuration strategy generating device further includes:

[0096] Acquire training samples of different sensitive content types, wherein the training samples are configured with sample labels corresponding to the review label configuration strategy;

[0097] Iteratively training a preset content review model based on each of the training samples and their corresponding sample labels;

[0098] Use the trained content review model to review the content to be reviewed and obtain the content review results.

[0099] The audit tag configuration strategy generating device further includes:

[0100] Get several different audit topics;

[0101] For any of the audit questions: repeatedly input the audit question into the preset question-answering model to obtain a plurality of first feedback information output by the preset question-answering model,

[0102] Sorting the first feedback information to train a reward model according to the sorting result;

[0103] For any of the audit questions: inputting the audit question into the preset question-answering model to obtain second feedback information output by the preset question-answering model;

[0104] Using the trained reward model to score the second feedback information;

[0105] Based on the scoring results and the review questions, the parameters of the preset question and answer model are adjusted.

[0106] The audit tag configuration strategy generating device further includes:

[0107] Obtaining a plurality of inquiry questions with different interception standards and audits, wherein each of the inquiry questions is configured with answer information;

[0108] Based on each of the inquiry questions and the corresponding answer information, parameters of the preset question-answering model are adjusted.

[0109] It should be noted here that the above-mentioned device provided in the embodiment of the present invention can implement all the method steps implemented in the above-mentioned method embodiment, and can achieve the same technical effect. The parts and beneficial effects that are the same as the method embodiment in this embodiment will not be described in detail here.

[0110] Figure 3 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 3 As shown, the electronic device may include: a processor 310, a memory 320, a communications interface 330 and a communications bus 340, wherein the processor 310, the memory 320 and the communications interface 330 communicate with each other through the communications bus 340. The processor 310 may call the logic instructions in the memory 320 to execute the audit label configuration strategy generation method, which includes: obtaining the interception standard document of the target customer and the audit standard specification document of the product; extracting multiple text information and multiple image information based on the audit standard specification document and the interception standard document; generating query instruction information based on each of the text information and each of the image information; inputting the query instruction information into the preset question and answer model to obtain the audit label configuration strategy output by the preset question and answer model.

[0111] In addition, the logic instructions in the above-mentioned memory 320 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0112] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the audit label configuration strategy generation method provided by the above-mentioned methods, the method comprising: obtaining an interception standard document of a target customer and an audit standard specification document of a product; extracting a plurality of text information and a plurality of image information based on the audit standard specification document and the interception standard document; generating query instruction information based on each of the text information and each of the image information; inputting the query instruction information into a preset question and answer model to obtain the audit label configuration strategy output by the preset question and answer model.

[0113] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the audit label configuration policy generation method provided by the above-mentioned methods.

[0114] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0115] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0116] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating an audit tag configuration strategy, characterized in that: include: Obtain the interception standard documents of target customers and the audit standard specification documents of products; Based on the audit standard specification document and the interception standard document, extract multiple text information and multiple image information; Generate inquiry instruction information based on each of the text information and each of the picture information; Inputting the query instruction information into a preset question-answering model to obtain an audit tag configuration strategy output by the preset question-answering model; The interception standard document and the audit standard specification document include multi-level label information, definition description information and sample information; wherein the multi-level label information represents the type label of the sensitive content, the definition description information represents the detailed description information of the type of sensitive content, and the sample information represents a detailed example or sample diagram; The extracting of multiple text information and multiple image information based on the audit standard specification document and the interception standard document includes: Performing text extraction and image extraction on the audit standard specification document and the interception standard document in the order of multi-level tag information, definition description information and sample information to obtain a plurality of text extraction information and image information associated with the text extraction information; Performing semantic segmentation processing on each of the text extraction information to obtain each of the text information; The generating of inquiry instruction information based on each of the text information and each of the picture information includes: Encoding each of the picture information to obtain picture encoding information; Based on the picture coding information and the text information associated therewith, generating inquiry instruction information in a preset inquiry format; Before inputting the query instruction information into a preset question-answering model to obtain the audit label configuration strategy output by the preset question-answering model, the method further includes: Get several different audit topics; For any of the audit questions: repeatedly input the audit question into the preset question-answering model to obtain a plurality of first feedback information output by the preset question-answering model, Sorting the first feedback information to train a reward model according to the sorting result; For any of the audit questions: inputting the audit question into the preset question-answering model to obtain second feedback information output by the preset question-answering model; Using the trained reward model to score the second feedback information; Based on the scoring results and the review questions, the parameters of the preset question and answer model are adjusted.

2. The audit tag configuration strategy generation method according to claim 1, characterized in that: After inputting the query instruction information into the preset question-answer model to obtain the audit tag configuration strategy output by the preset question-answer model, the method further includes: Acquire training samples of different sensitive content types, wherein the training samples are configured with sample labels corresponding to the review label configuration strategy; Iteratively training a preset content review model based on each of the training samples and their corresponding sample labels; Use the trained content review model to review the content to be reviewed and obtain the content review results.

3. The audit tag configuration strategy generation method according to claim 1, characterized in that: Before inputting the query instruction information into a preset question-answering model to obtain the audit label configuration strategy output by the preset question-answering model, the method further includes: Obtaining a plurality of inquiry questions with different interception standards and audits, wherein each of the inquiry questions is configured with answer information; Based on each of the inquiry questions and the corresponding answer information, parameters of the preset question-answering model are adjusted.

4. A device for generating an audit tag configuration strategy, characterized in that: include: The acquisition module is used to obtain the interception standard documents of target customers and the audit standard specification documents of products; An extraction module, used for extracting a plurality of text information and a plurality of image information based on the audit standard specification document and the interception standard document; A generating module, used for generating inquiry instruction information based on each of the text information and each of the picture information; A label configuration module, used to input the query instruction information into a preset question-answering model to obtain an audit label configuration strategy output by the preset question-answering model; The extracting of multiple text information and multiple image information based on the audit standard specification document and the interception standard document includes: Performing text extraction and image extraction on the audit standard specification document and the interception standard document in the order of multi-level tag information, definition description information and sample information to obtain a plurality of text extraction information and image information associated with the text extraction information; Performing semantic segmentation processing on each of the text extraction information to obtain each of the text information; The generating of query instruction information based on each of the text information and each of the picture information includes: Encoding each of the picture information to obtain picture encoding information; Based on the picture coding information and the text information associated therewith, generating inquiry instruction information in a preset inquiry format; The interception standard document and the audit standard specification document include multi-level label information, definition description information and sample information; wherein the multi-level label information represents the type label of the sensitive content, the definition description information represents the detailed description information of the type of sensitive content, and the sample information represents a detailed example or sample diagram; Before inputting the query instruction information into a preset question-answering model to obtain the audit label configuration strategy output by the preset question-answering model, the method further includes: Get several different audit topics; For any of the audit questions: repeatedly input the audit question into the preset question-answering model to obtain a plurality of first feedback information output by the preset question-answering model, Sorting the first feedback information to train a reward model according to the sorting result; For any of the audit questions: inputting the audit question into the preset question-answering model to obtain second feedback information output by the preset question-answering model; Using the trained reward model to score the second feedback information; Based on the scoring results and the review questions, the parameters of the preset question and answer model are adjusted.

5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the program, the audit tag configuration strategy generation method according to any one of claims 1 to 3 is implemented.

6. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the audit tag configuration strategy generation method according to any one of claims 1 to 3 is implemented.

Images