Content auditing method, system, device and equipment, storage medium and program product

By generating audit prompt words and using generative large models for content review, the problem of low content review efficiency in the existing technology is solved, efficient and accurate audit results are achieved, and the audit needs of complex content such as the financial field are applicable.

CN119005183APending Publication Date: 2024-11-22CHINA CONSTRUCTION BANK +1
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411105466.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-13
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

In the existing technology, the content review efficiency is low and it is impossible to effectively handle various types of published content, especially in the review of advertising, activities and publicity posters in the financial field. The auditors need to interpret complex rules and regulations, which makes the review difficult and Inefficient.

Method used

By obtaining the review process information and normative information of the target published content, input it into the preset prompt word generation model to generate audit prompt words, and review the content based on these prompt words, and use the generative big model to train and adjust the parameter , improve audit efficiency and accuracy.

Benefits of technology

It greatly reduces the work of auditors to interpret standardized audit information and processes, improves audit efficiency, ensures compliance and accuracy of audit results, and is suitable for various types of publishing content and complex business processes.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119005183A_ABST
    Figure CN119005183A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of artificial intelligence, and particularly discloses a content auditing method, system, device and equipment, a storage medium and a program product. The method comprises the following steps: in response to an auditing request of target published content, obtaining multiple pieces of auditing process information and auditing normativity information of the target published content; inputting each piece of auditing process information and auditing normativity information into a preset prompt word generation model, and generating auditing prompt words of a plurality of auditing processes; and according to the auditing prompt word of each auditing process, auditing the target published content to obtain an auditing result of the target published content. By adopting the method, the auditing efficiency of the published content can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a content review method, system, device, equipment, storage medium and program product. Background Art

[0002] With the widespread application of Internet technology, more and more information is published and disseminated through the Internet. Taking the financial field as an example, with the rapid development of the financial industry, the types and complexity of financial products are increasing. In order to recommend financial products to customers, some advertisements, activities, text messages, promotional posters and other content related to financial products are usually released.

[0003] In related technologies, in order to ensure that the published content complies with relevant rules and regulations and avoid violations, it is necessary to review the content before publishing it to detect whether there is any information that violates the regulations.

[0004] However, the method of reviewing published content in related technologies has a technical problem of low review efficiency. Summary of the invention

[0005] Based on this, it is necessary to provide a content review method, system, device, equipment, storage medium and program product to address the above technical problems, which can improve the efficiency of reviewing published content.

[0006] In a first aspect, an embodiment of the present application provides a content review method, comprising:

[0007] In response to a review request for target published content, obtaining multiple review process information and review normative information of the target published content;

[0008] Input each audit process information and audit normative information into a preset prompt word generation model to generate audit prompt words for multiple audit processes;

[0009] According to the audit prompt words of each audit process, the target published content is audited to obtain the audit result of the target published content.

[0010] In one embodiment, each audit process information and audit normative information is input into a preset prompt word generation model to generate audit prompt words for multiple audit processes, including:

[0011] Obtain the prompt word generation instructions for each review process;

[0012] Input each prompt word generation instruction, each audit process information and audit normative information into the prompt word generation model to obtain the audit prompt words of each audit process.

[0013] In one embodiment, the target published content is reviewed according to the review prompt words of each review process to obtain the review result of the target published content, including:

[0014] Perform vector conversion processing on the multimodal data in the target published content to obtain a high-dimensional vector of the multimodal data;

[0015] According to the high-dimensional vector and each audit prompt word, the target published content is audited to obtain the audit result.

[0016] In one embodiment, before performing vector conversion processing on the multimodal data in the target published content to obtain a high-dimensional vector of the multimodal data, the method further includes:

[0017] The multimodal data is preprocessed to obtain preprocessed multimodal data; the preprocessing method includes at least one of data cleaning processing, word segmentation processing and syntactic analysis processing.

[0018] In one embodiment, the target published content is audited according to the high-dimensional vector and each audit prompt word to obtain the audit result, including:

[0019] According to the high-dimensional vector, multiple similar published contents and historical review results of each similar published content are obtained from the historical review database;

[0020] Using similar published content and historical review results, the preset generative large model is trained to obtain a published content review model;

[0021] According to the audit prompt words and the published content audit model, the target published content is audited to obtain the audit results.

[0022] In one embodiment, similar published content and historical review results are used to train a preset generative large model to obtain a published content review model, including:

[0023] Obtain historical standard business data for each similar published content;

[0024] Input each historical standard business data, each similar published content and each audit prompt word into the generative big model to obtain the audit prediction information output by the generative big model;

[0025] According to the audit prediction information and historical audit results, the parameters of the generative big model are adjusted to obtain the published content audit model.

[0026] In one embodiment, according to the audit prediction information and each historical audit result, the parameters of the generative big model are adjusted to obtain a published content audit model, including:

[0027] Determine the predicted loss value based on the audit prediction information, historical audit results and the preset loss function;

[0028] Based on the predicted loss value, the model parameters of the generative large model are adjusted until the predicted loss value is less than the preset threshold, and the published content review model is obtained.

[0029] In one embodiment, the target published content is reviewed according to the review prompt words and the published content review model to obtain the review results, including:

[0030] Obtain standard business data of target published content;

[0031] Input standard business data, target publication content and various review prompts into the publication content review model to obtain the review results.

[0032] In one embodiment, multiple review process information and review normative information of target published content are obtained, including:

[0033] Obtain the review guidance documents and review specification documents for published content review;

[0034] According to the audit guidance documents, extract the information of each audit process, and according to the audit specification documents, extract the audit normative information.

[0035] In a second aspect, the embodiment of the present application further provides a content review system, which includes a review model construction module, a review prompt word generation module, and a review result generation module; the review prompt word generation module is respectively connected to the review model construction module and the review result generation module in communication;

[0036] An audit model building module is used to obtain multiple similar published contents and historical audit results of each similar published content according to the multimodal data in the target published content, and use each similar published content and each historical audit result to perform model training to obtain a published content audit model;

[0037] An audit prompt word generation module is used to process multiple audit process information and audit normative information of target published content using a preset prompt word generation model to generate audit prompt words for multiple audit processes;

[0038] The audit result generation module is used to process the target published content and each audit prompt word using the published content audit model to obtain the audit result of the target published content.

[0039] In a third aspect, the present application also provides a content review device, including:

[0040] An information acquisition module, used to respond to a review request for a target published content and acquire multiple review process information and review normative information of the target published content;

[0041] A prompt word generation module, used to input each audit process information and audit normative information into a preset prompt word generation model to generate audit prompt words for multiple audit processes;

[0042] The content review module is used to review the target published content according to the review prompt words of each review process and obtain the review result of the target published content.

[0043] In a fourth aspect, an embodiment of the present application further provides a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps in any one of the embodiments of the first aspect are implemented.

[0044] In a fifth aspect, the embodiments of the present application further provide a computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps in any one of the embodiments of the first aspect are implemented.

[0045] In a sixth aspect, the embodiments of the present application further provide a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps in any one of the embodiments in the first aspect are implemented.

[0046] The above-mentioned content review method, system, device, equipment, storage medium and program product obtain multiple review process information and review normative information of the target published content by responding to the review request of the target published content, and then input each review process information and review normative information into the preset prompt word generation model to generate multiple review prompt words of the review process, and then review the target published content according to the review prompt words of each review process to obtain the review result of the target published content. In this method, when reviewing the target published content, the review is based on the review prompt words of each review process, and the review prompt words of each review process are generated by inputting multiple review process information and review normative information of the target published content into the prompt word generation model, which is equivalent to, when reviewing the published content, the review prompt words can be extracted from the review process information and the review normative information to conduct content review, which greatly reduces the work of the reviewer to interpret the review normative information and the review process, and improves the review efficiency. In addition, when generating audit prompt words, multiple audit process information is also input into the prompt word generation model. This is equivalent to using the model to learn each step of the audit process on the basis of the existing audit process, and generating audit prompt words to assist in content review. In this way, by performing content review without modifying the original audit business process, compliance with audit requirements is ensured, and the accuracy of the audit results is also guaranteed. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0048] FIG1 is a diagram of an application environment of a content audit method in one embodiment;

[0049] FIG2 is a schematic diagram of a flow chart of a content review method in one embodiment;

[0050] FIG3 is a schematic diagram of a process for generating audit prompt words in one embodiment;

[0051] FIG4 is a schematic diagram of a process for determining an audit result in one embodiment;

[0052] FIG5 is a schematic diagram of a process for determining an audit result in another embodiment;

[0053] FIG6 is a schematic diagram of a process for determining a published content review model in one embodiment;

[0054] FIG7 is a schematic diagram of a process for determining an audit result in another embodiment;

[0055] FIG8 is a schematic diagram of a process for obtaining multiple audit process information and audit normative information in one embodiment;

[0056] FIG9 is a flow chart of a content review method in another embodiment;

[0057] FIG10 is a schematic diagram of the structure of a content review device in one embodiment;

[0058] FIG. 11 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0060] The content audit method provided in the embodiment of the present application can be applied in the application environment shown in Figure 1. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. Optionally, the terminal 102 is used to receive the audit results of the target published content. When the server 104 audits the target published content and obtains the audit results of the target published content, the audit results of the target published content can be sent to the terminal 102. Among them, the terminal 102 can be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The server 104 can be implemented with an independent server or a server cluster consisting of multiple servers.

[0061] The technical background of this application is described below.

[0062] With the widespread application of Internet technology, more and more information is published and disseminated through the Internet.

[0063] Take the financial sector as an example. With the rapid development of the financial industry, the types and complexity of financial products are increasing. In order to recommend financial products to customers, some advertisements, activities, text messages, promotional posters and other content related to financial products are usually released. The relevant review departments have increasingly stringent requirements for the review of published content, and have formulated a number of rules and regulations. The workload of content review staff has increased year by year, and they are faced with the challenges of complex interpretation of rules and regulations and difficult review. Therefore, it is urgent to adopt artificial intelligence technology to reduce the work pressure of business personnel and improve the efficiency and accuracy of review.

[0064] In related technologies, artificial intelligence technology in other fields has been widely used in the review of documents and pictures. One method is to obtain the review identification code of the image to be tested, perform similarity comparison, and reduce the cost of repeated reviews. However, this review method only processes images, and its application scope is limited to image review, and it cannot cover other types of review tasks; it lacks support for the review of text, complex documents, and various business processes, and cannot meet the diverse needs of different fields. Another method is to match question statements through the contextual semantic vector representation of text sentences to improve the efficiency of file review. However, this review method cannot learn from the experience of manual review experts, and the intelligent review capability cannot be improved over time and with the accumulation of experience; and this method relies on the domain knowledge graph. If the knowledge graph is not perfect or updated in time, it may lead to inaccurate review results; in addition, there are certain limitations for review tasks in specific fields, and it is difficult to apply to a variety of review processes.

[0065] Based on this, the embodiment of the present application provides a content review method. When reviewing the target published content, the review is conducted based on the review prompt words of each review process, and the review prompt words of each review process are generated by inputting multiple review process information and review normative information of the target published content into the prompt word generation model, which is equivalent to, when reviewing the published content, the review prompt words can be extracted from the review process information and the review normative information to conduct content review, which greatly reduces the work of the reviewer to interpret the review normative information and the review process, and improves the review efficiency. In addition, when generating the review prompt words, multiple review process information is also input into the prompt word generation model, which is equivalent to, on the basis of the existing review process, using the model to learn each step of the review process and generate review prompt words to assist in content review. In this way, by conducting content review without modifying the original review business process, compliance with the review requirements is ensured, and the accuracy of the review results is also guaranteed.

[0066] It should be noted that the beneficial effects brought about by the embodiments of the present application or the technical problems solved are not limited to this one, but may also include other implicit or related problems. For details, please refer to the description of the following embodiments.

[0067] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0068] In an exemplary embodiment, as shown in FIG2 , a content audit method is provided, which is described by taking the method applied to the server 104 in FIG1 as an example, and includes the following steps 201 to 203. Among them:

[0069] S201, in response to a review request for target published content, obtaining a plurality of review process information and review normative information of the target published content.

[0070] Published content refers to advertisements, activities, text messages, promotional posters, information, articles, etc. that are published on specific platforms or media for users to browse, read or watch. Target published content refers to published content that requires content review. The content presentation form of published content may include but is not limited to text, pictures, videos, voice, etc.

[0071] The review request of the target published content refers to the request triggered when the target published content needs to be reviewed in the content review application of the server. Naturally, after receiving the review request, the server needs to conduct content review on the target published content to detect whether there is any information that violates relevant rules and regulations.

[0072] In one embodiment, the request is triggered by the target user through the server interface in the content review application; wherein, the triggering method includes but is not limited to external input device clicks, voice control, touch screen triggering, etc., and the embodiment of the present application does not limit the triggering method.

[0073] In another embodiment, the request is triggered according to a preset automatic program; for example, if the server receives the target published content uploaded by the target user, the review request for the target published content is automatically triggered according to the preset program; or, after the server receives the target published content uploaded by the target user, if the number of received target published content reaches a certain threshold, and / or, when the application is in an idle state, the review request for the target published content is automatically triggered according to the preset program. The embodiment of the present application does not limit the triggering method of the content review request.

[0074] In the embodiment of the present application, the review process information refers to each step of reviewing the published content, as well as the information and precautions that need to be reviewed in each step. Review normative information refers to information such as regulations, guidelines, industry specifications, regulations, rules and regulations related to the review of published content, which are formulated for the review of published content.

[0075] Usually, the relevant review department has already designed and stipulated in advance the process, steps, precautions and other information for the content review work. When conducting the review operation, the content review can be carried out according to the pre-set process and steps.

[0076] Exemplarily, the method of obtaining multiple audit process information of the target published content may be to obtain the audit process file related to the published content, and then extract the relevant information of each audit step from the audit process file, and determine the relevant information of each audit step as multiple audit process information. Among them, the audit process file refers to a file generated based on the steps and procedures pre-designed by the relevant audit department for the audit work of the published content, including each step of the audit and the information and precautions required for each step.

[0077] Exemplarily, the method of obtaining the audit normative information of the target published content can be to obtain the audit normative document related to the published content, and then extract relevant regulations, criteria, industry norms, regulations, rules and regulations and other information from the audit normative document, and then determine the extracted regulations, criteria, industry norms, regulations, rules and regulations and other information as the audit normative information. Among them, the audit normative document is a document formulated by the relevant department for the compliance of the published content, which includes the formulated rules, regulations, industry norms, regulations, criteria and other information.

[0078] S202: Input each audit process information and audit normative information into a preset prompt word generation model to generate audit prompt words for multiple audit processes.

[0079] The prompt word generation model is a technology that uses natural language processing (NLP) technology to generate prompt words related to specific topics or needs by analyzing text information. In the embodiment of the present application, the preset prompt word generation model can be a generative large model.

[0080] Prompt words play an important role in the field of artificial intelligence (AI) and machine learning. They are instructions or inputs used to connect AI models with user needs and provide a "bridge" for services. Prompt words can be questions, descriptions, commands, etc., or even dialogues or stories, in various forms. The main purpose of prompt words is to tell the AI ​​model what you want, so as to help the AI ​​model understand and respond accordingly or generate creative content. In the embodiment of the present application, the review prompt words refer to the key words and phrases input into the review model when reviewing the published content.

[0081] In one embodiment, a prompt word generation model is used to generate audit prompt words for multiple audit processes by inputting each audit process information and audit normative information into a preset prompt word generation model, and the prompt word generation model performs keyword extraction, logical association and other processing on the input audit process information and audit normative information. After that, the prompt word generation model can generate audit prompt words for multiple audit processes based on the processed information and the prompt word generation principles customized by the prompt word generation model, and output the audit prompt words for multiple audit processes.

[0082] In another embodiment, a prompt word generation model is used to generate audit prompt words for multiple audit processes by obtaining prompt word generation requirements, and then inputting the prompt word generation requirements, each audit process information and audit normative information into a preset prompt word generation model, and the prompt word generation model analyzes each audit process information and audit normative information, and generates audit prompt words for multiple audit processes based on the prompt word generation requirements, and outputs audit prompt words for multiple audit processes. In the embodiment of the present application, the audit prompt words may include key review points, referenced content in the audit normative information, precautions and other information.

[0083] S203, reviewing the target published content according to the review prompt words of each review process, and obtaining the review result of the target published content.

[0084] In one embodiment, the target published content is reviewed according to the review prompt words of each review process, and the method for obtaining the review result of the target published content can be to extract multiple key sentences in the target published content, and then according to the established rules, the multiple key sentences are reviewed according to the review prompt words of each review process, and the review result of each key sentence is obtained, and the review result of each key sentence is determined as the review result of the target published content.

[0085] In another embodiment, the target published content is reviewed according to the review prompts of each review process, and a method for obtaining the review result of the target published content can be to input the review prompts of each review process and the target published content into a pre-trained content review model, and the content review model processes the target published content and the review prompts of each review process to obtain the review result of the target published content.

[0086] It should be noted that in the embodiment of the present application, when the target published content is presented in the form of video, voice or picture, the video, voice or picture can be converted into text form before the review operation is performed. In addition, it should be noted that if the published content in the application embodiment is content that is open to the Internet and shared to each user, such as news reports, social media posts, etc., then the review of the published content can be to review whether there are grammatical errors, factual errors, or bad information, sensitive information, etc. If the published content in the embodiment of the present application is content published to consumers, such as advertisements, text messages, posters, etc., then the review of the published content can be to review whether it contains false content, misleading content, misleading content for consumers, or content that does not comply with consumer protection rights, or to review whether it includes relevant content on consumer rights protection that should be written.

[0087] In the content review method provided in the embodiment of the present application, by responding to the review request of the target published content, multiple review process information and review normative information of the target published content are obtained, and then each review process information and review normative information are input into the preset prompt word generation model to generate multiple review process review prompt words, and then the target published content is reviewed according to the review prompt words of each review process to obtain the review result of the target published content. In this method, when reviewing the target published content, the review is based on the review prompt words of each review process, and the review prompt words of each review process are generated by inputting multiple review process information and review normative information of the target published content into the prompt word generation model, which is equivalent to, when reviewing the published content, the review prompt words can be extracted from the review process information and the review normative information to conduct content review, which greatly reduces the work of the reviewer to interpret the review normative information and the review process, and improves the review efficiency. In addition, when generating audit prompt words, multiple audit process information is also input into the prompt word generation model. This is equivalent to using the model to learn each step of the audit process on the basis of the existing audit process, and generating audit prompt words to assist in content review. In this way, by performing content review without modifying the original audit business process, compliance with audit requirements is ensured, and the accuracy of the audit results is also guaranteed.

[0088] Based on the above embodiment, an embodiment is provided to illustrate the process of inputting each audit process information and audit normative information into a preset prompt word generation model to generate audit prompt words for multiple audit processes.

[0089] In an exemplary embodiment, as shown in FIG3 , each audit process information and audit normative information are input into a preset prompt word generation model to generate audit prompt words for multiple audit processes, including:

[0090] S301, obtaining prompt word generation instructions for each review process.

[0091] The prompt word generation instruction is a related instruction for guiding or stimulating the model to generate prompt words. In the embodiment of the present application, each audit process corresponds to a prompt word generation instruction, so that for any audit process, the audit prompt words of the audit process can be generated according to the prompt word generation instruction of the audit process.

[0092] In one embodiment, the method of obtaining the prompt word generation instruction for each audit process can be through a user interface, and the content input in the user interface is obtained as the prompt word generation instruction for each audit process. For example, when it is necessary to obtain the prompt word generation instruction for each audit process, the instruction input interface is displayed to the user, wherein the instruction input interface includes instruction filling areas for different audit processes, and the user can fill in the prompt word generation instruction for each audit process in each instruction filling area in turn, and click to confirm, so that the server can receive the prompt word generation instruction for each audit process filled in by the user, and then determine the prompt word generation instruction for each audit process filled in by the user as the prompt word generation instruction for each audit process.

[0093] In another embodiment, the method of obtaining the prompt word generation instructions for each audit process can be to obtain multiple historical prompt word generation instructions from the historical instruction library as the prompt word generation instructions for each audit process. For example, the historical instruction library includes multiple historical prompt word generation instructions input into the prompt word generation model when generating prompt words at historical time. Multiple historical prompt word generation instructions can be selected from the historical instruction library to be displayed to the user, and the user confirms whether to determine the multiple historical prompt word generation instructions as the prompt word generation instructions for each audit process. If the user confirms that the selected multiple historical prompt word generation instructions can be used as the prompt word generation instructions for each audit process, then the multiple historical prompt word generation instructions are the prompt word generation instructions for each audit process; if the user believes that the selected multiple historical prompt word generation instructions do not meet the user's needs, other historical prompt word generation instructions can also be selected from the historical instruction library as the prompt word generation instructions for each audit process. It should be noted that in the embodiment of the present application, the prompt word generation instructions for each audit process can be the same or different, and the embodiment of the present application does not limit this, and can be determined according to actual needs.

[0094] S302, inputting each prompt word generation instruction, each audit process information and audit normative information into the prompt word generation model to obtain the audit prompt words of each audit process.

[0095] After obtaining the prompt word generation instructions for each audit process, the prompt word generation instructions for each audit process, each audit process information, and audit normative information are input into the prompt word generation model, and the prompt word generation model analyzes and processes each prompt word generation instruction, each audit process information, and audit normative information to generate audit prompt words for each audit process. The processing method of the prompt word generation model may include keyword extraction, context analysis, and logical association.

[0096] In the embodiment of the present application, take the prompt word generation instruction as "I am in the process of publishing content review, and I need to use the review prompt words of each review process to review the published content to improve the review efficiency. Please help me generate the review prompt words for each review process. Among them, the review prompt words need to include the key review points of each review process and the reference content of the review normative information." As an example, the instruction and each review process information and review normative information are input into the prompt word generation model, and the prompt word generation model can analyze the content in the instruction, and perform operations and processing such as keyword extraction on each review process information and review normative information to generate the review prompt words for each review process. It should be noted that the above-mentioned prompt word generation instruction is only an example. The embodiment of the present application does not limit the content in the prompt word generation instruction, as long as the prompt word generation instruction can accurately describe the goals and requirements. Of course, in order to make the generated review prompt words more in line with user needs, specific examples can also be provided in the prompt word generation instruction.

[0097] In the content review method provided in the embodiment of the present application, by obtaining the prompt word generation instructions of each review process, and then inputting each prompt word generation instruction, each review process information and review normative information into the prompt word generation model, the review prompt words of each review process are obtained. In this method, by introducing the prompt word generation instructions of each review process, the specific tasks that the prompt word generation model should perform and the specific outputs generated are clarified, so that the prompt word generation model can output more expected review prompt words based on different requirements, improve the quality and accuracy of the review prompt words, and lay the foundation for obtaining accurate review results of the target published content.

[0098] Based on any of the above embodiments, an embodiment is provided to illustrate the process of reviewing the target published content according to the review prompt words of each review process and obtaining the review result of the target published content.

[0099] In an exemplary embodiment, as shown in FIG4 , the target published content is reviewed according to the review prompt words of each review process to obtain the review result of the target published content, including:

[0100] S401, performing vector conversion processing on the multimodal data in the target published content to obtain a high-dimensional vector of the multimodal data.

[0101] Multimodal data refers to data in different presentation forms in the target published content, such as text data, image data, video data, voice data, etc.

[0102] Exemplarily, performing vector conversion processing on the multimodal data in the target published content to obtain a high-dimensional vector of the multimodal data can be performed by performing vector conversion on the text data in the multimodal data to obtain a text high-dimensional vector, and performing vector conversion on non-text data such as pictures and videos in the multimodal data to obtain a non-text high-dimensional vector, and then integrating the text high-dimensional vector and the non-text high-dimensional vector to obtain a high-dimensional vector of the multimodal data.

[0103] Among them, the methods of vectorizing text data may include unique hot encoding, bag-of-words model, word frequency-inverse document frequency, word-vector model and other processing methods. The method of vectorizing images may be implemented by using image processing technology, such as convolutional neural network in deep learning. The method of vectorizing videos may be implemented by using video vectorization technology, including extracting key frames from videos, extracting features from each frame of video, as well as local feature extraction and global feature extraction. The integration of text high-dimensional vectors and non-text high-dimensional vectors may be implemented by using average pooling technology or concat technology in deep learning.

[0104] In order to improve the quality, availability and applicability of the data and to ensure the accuracy and effectiveness of subsequent analysis and modeling, the multimodal data are preprocessed before vector transformation.

[0105] In an exemplary embodiment, before performing vector conversion processing on the multimodal data in the target published content to obtain a high-dimensional vector of the multimodal data, the method further includes:

[0106] The multimodal data is preprocessed to obtain preprocessed multimodal data; the preprocessing method includes at least one of data cleaning processing, word segmentation processing and syntactic analysis processing.

[0107] There may be errors, missing values, inconsistent formats and other problems in multimodal data. You can first perform data cleaning on the multimodal data. For the text data in the multimodal data, after data cleaning, you can also perform word segmentation and syntactic analysis on the text data to extract key sentences and context information, paving the way for subsequent high-dimensional vector conversion.

[0108] In this embodiment, before the multimodal data in the target published content is subjected to vector conversion processing, the multimodal data is preprocessed to obtain preprocessed multimodal data; wherein the preprocessing method includes at least one of data cleaning processing, word segmentation processing and syntactic analysis processing. In this method, in order to enable subsequent tasks to be carried out more accurately and efficiently, the multimodal data is cleaned to remove irrelevant and erroneous information, thereby improving the quality and reliability of the data and providing a better foundation for subsequent analysis and modeling. In addition, the multimodal data is subjected to word segmentation processing and syntactic analysis processing to extract key sentences and context information, which provides convenience and foundation for subsequent vector conversion.

[0109] S402, review the target published content according to the high-dimensional vector and each review prompt word to obtain a review result.

[0110] After obtaining the high-dimensional vector of the multimodal data, the high-dimensional vector and each review prompt word may be used to review the target published content to obtain the review result of the target published content.

[0111] Exemplarily, the method of reviewing the target published content according to the high-dimensional vector and each review prompt word can be to input the high-dimensional vector, each review prompt word and the target published content into a pre-trained review model, and the review model processes the high-dimensional vector, each review prompt word and the target published content to obtain the review result of the target published content. Among them, the review model can be obtained by training a generative large model using each review prompt word, a large amount of historical published content and the high-dimensional vector corresponding to each historical published content.

[0112] In this embodiment, the target published content includes multimodal data, and multiple forms of published content can be reviewed at the same time to meet the diverse review needs in various fields.

[0113] In the content review method provided in the embodiment of the present application, a high-dimensional vector of the multimodal data is obtained by performing vector conversion processing on the multimodal data in the target published content, and then the target published content is reviewed according to the high-dimensional vector and each review prompt word to obtain the review result. In this method, a high-dimensional vector of the multimodal data is obtained by performing vector conversion processing on the multimodal data in the target published content, so that various data can be compared and analyzed in the same dimensional space, etc., laying the foundation for subsequent vectorized search.

[0114] Based on the above embodiment, an embodiment is provided to illustrate the process of reviewing the target published content according to the high-dimensional vector and each review prompt word and obtaining the review result.

[0115] In an exemplary embodiment, as shown in FIG5 , the target published content is audited according to the high-dimensional vector and each audit prompt word to obtain an audit result, including:

[0116] S501, obtaining multiple similar published contents and historical review results of each similar published content from a historical review database according to the high-dimensional vector.

[0117] The historical review database refers to a database that stores historical published content. The historical review database includes multiple historical published content, historical review results of each historical published content, and historical standard business data of each historical published content. The historical business data refers to the attribute information of each historical published content, such as the publishing form, business type, publishing time, etc. of the historical published content.

[0118] In the embodiment of the present application, a vectorized search is performed based on a high-dimensional vector of the target published content to quickly and accurately retrieve historically similar published content.

[0119] Exemplarily, according to the high-dimensional vector, a method of obtaining multiple similar published contents from the historical audit database may be to obtain each historical published content from the historical audit database, and perform vector conversion processing on the data in each historical published content to obtain the historical high-dimensional vector of each historical published content, and then calculate the vector similarity between the high-dimensional vector of the target published content and each historical high-dimensional vector, and determine the similar published contents according to each vector similarity. For example, each vector similarity may be compared with a preset similarity threshold, and the historical published contents corresponding to the vector similarity greater than the similarity threshold are determined as multiple similar published contents.

[0120] After obtaining similar published contents, historical review results of each similar published content are obtained from a historical review database.

[0121] S502, using each similar published content and each historical review result, the preset generative large model is trained to obtain a published content review model.

[0122] Model training is performed based on the multiple similar published contents obtained above and the historical review results of each similar published content to obtain a published content review model.

[0123] In one embodiment, multiple similar published contents and each review prompt word are used as model inputs, and the historical review results of each similar published content are used as targets. Multiple similar published contents and each review prompt word are input into a preset generative big model to obtain the prediction results output by the generative big model. Then, based on the prediction results and the historical review results as the targets, the parameters of the generative big model are adjusted to obtain a published content review model.

[0124] In another embodiment, multiple similar published contents and historical review results of each similar published content are divided into data sets to obtain a training data set and a test data set. The training data set and each review prompt word are used to train a preset generative large model to obtain an initial model. The initial model is then tested using the test data set and each review prompt word. If the test passes, the initial model is determined as a content review model. If the test fails, another portion of similar published content is obtained, and these data are used to adjust parameters of the initial model until the test passes to obtain a published content review model.

[0125] S503, auditing the target published content according to each audit prompt word and the published content audit model to obtain the audit result.

[0126] After the published content review model is obtained through training, the target published content is reviewed according to the published content review model and various review prompts to obtain the review results.

[0127] Exemplarily, according to the review prompt words and the published content review model, the method for reviewing the target published content can be to input each review prompt word and the target published content into the published review model, and the published review model processes each review prompt word and the target published content to obtain the review result of the target published content.

[0128] In the content review method provided in the embodiment of the present application, based on the high-dimensional vector, multiple similar published contents and the historical review results of each similar published content are obtained from the historical review database, and then each similar published content and each historical review result are used to perform model training on the preset generative large model to obtain a published content review model, and then the target published content is reviewed according to each review prompt word and the published content review model to obtain the review result. In this method, through the high-dimensional vector, a vectorized search is performed in the historical review database, and multiple similar published contents can be quickly and accurately retrieved, providing data support for training the published content review model, and improving the reference and scientific nature of the review decision. Then, based on the obtained multiple similar published contents and the historical review results of each similar published content, the model is trained to obtain a published content review model, so that the model can learn from historical review experience and improve the review accuracy of the published content review model.

[0129] Based on the above embodiment, an embodiment is provided to illustrate the process of using similar published contents and historical review results to train the preset generative big model to obtain a published content review model.

[0130] In an exemplary embodiment, as shown in FIG6 , similar published contents and historical audit results are used to train a preset generative large model to obtain a published content audit model, including:

[0131] S601, obtaining historical standard business data of each similar published content.

[0132] As mentioned above, the historical review database includes historical standard business data of multiple historical published contents, and the historical standard business data of each similar published content can be obtained from the historical review database.

[0133] S602, input each historical standard business data, each similar published content and each audit prompt word into the generative big model to obtain the audit prediction information output by the generative big model.

[0134] In the embodiment of the present application, the review prediction information refers to the prediction information of the review results of each similar published content by the generative big model.

[0135] Based on the historical standard business data of each similar published content obtained above, the generative large model is trained.

[0136] After obtaining the historical standard business data of each similar published content, each historical standard business data, each similar published content and each audit prompt word are input into the generative big model, and the generative big model processes each historical standard business data, each similar published content and each audit prompt word, and outputs the audit prediction information.

[0137] S603, according to the audit prediction information and historical audit results, the parameters of the generative big model are adjusted to obtain the published content audit model.

[0138] Based on the review prediction information output by the above-mentioned generative big model, the parameters of the generative big model are adjusted to obtain the published content review model.

[0139] After obtaining the audit prediction information output by the generative large model, the model parameters of the generative large model can be adjusted according to the loss value between the audit prediction information and each historical audit result.

[0140] Based on this, in one embodiment, the predicted loss value is determined according to the audit prediction information, various historical audit results and a preset loss function; based on the predicted loss value, the model parameters of the generative large model are adjusted until the predicted loss value is less than the preset threshold, thereby obtaining a published content audit model.

[0141] In the embodiment of the present application, a pre-set loss function is obtained, and then the audit prediction information and each historical audit result are substituted into the loss function to obtain a predicted loss value. Then, according to the predicted loss value, the model parameters of the generative large model are adjusted, and then the generative large model with adjusted parameters is trained again, and the predicted loss value is calculated again. The model parameters of the generative large model are adjusted continuously based on the predicted loss value until the predicted loss value is less than the preset threshold, the parameter adjustment is stopped, and the current generative large model is determined as the published content audit model.

[0142] In this embodiment, the predicted loss value is determined based on the audit prediction information, each historical audit result and the preset loss function, and then based on the predicted loss value, the model parameters of the generative large model are adjusted until the predicted loss value is less than the preset threshold value, thereby obtaining the published content audit model. In this method, by calculating the predicted loss value based on the audit prediction information and each historical audit result, and adjusting the model parameters of the generative large model based on the predicted loss value, the performance of the generative large model is optimized, so that the model can better approximate the real result, and the prediction accuracy of the model is improved.

[0143] In the content review method provided in the embodiment of the present application, by obtaining the historical standard business data of each similar published content, and then inputting each historical standard business data, each similar published content and each review prompt word into the generative big model, the review prediction information output by the generative big model is obtained, and then the parameters of the generative big model are adjusted according to the review prediction information and each historical review result to obtain the published content review model. In this method, by inputting each historical standard business data, each similar published content and each review prompt word into the generative big model for model training, the generative big model can learn and accumulate the review experience of multiple similar published contents, and the intelligent review capability is continuously improved with time and the increase of data, thereby improving the review accuracy of the published content review model.

[0144] Based on the above embodiment, an embodiment is provided to illustrate the process of auditing the target published content according to each audit prompt word and the published content audit model and obtaining the audit result.

[0145] In an exemplary embodiment, as shown in FIG7 , the target published content is reviewed according to each review prompt word and the published content review model to obtain the review result, including:

[0146] S701, obtaining standard business data of target published content.

[0147] When training the published content review model, the historical standard business data of multiple similar published contents are also input into the generative large model for learning. In the process of applying the published content review model, the standard business data of the target published content also needs to be input into the published content review model.

[0148] In one embodiment, the method of obtaining the standard business data of the target published content can be to obtain it from the content input in the user interface. For example, when it is necessary to obtain the standard business data of the target published content, the business data filling interface is displayed to the user, wherein the business data filling interface includes options of the same data type as the historical standard business data, for example, the historical standard business data includes the release time and the release form, and the business data filling interface also includes the release time and the release form options, and the user can enter the corresponding data through the input box, selection box, etc. corresponding to each option, and click OK, so that the server can obtain the data entered by the user and determine the data entered by the user as the standard business data of the target published content.

[0149] In another embodiment, the standard business data of the target published content may be obtained from a business data table. For example, the mapping relationship between different published content and standard business data is stored in the business data table. When the standard business data of the target published content needs to be obtained, the corresponding standard business data can be matched from the above mapping relationship in the business data table according to the target published content, and the matched standard business data is determined as the standard business data of the target published content.

[0150] S702, input the standard business data, target published content and various review prompt words into the published content review model to obtain the review result.

[0151] After obtaining the target business data, the standard business data, target publication content and various review prompts can be input into the publication content review model to obtain the review results output by the publication content review model.

[0152] In the content review method provided in the embodiment of the present application, the standard business data of the target published content is obtained, and then the standard business data, the target published content and each review prompt word are input into the published content review model to obtain the review result. In this method, when the target published content is reviewed according to the published content review model, the standard business data of the target published content is obtained, and then the standard business data, the target published content and each review prompt word are input into the published content review model to obtain the review result, so that the input data in the application process of the published content review model is consistent with the input data in the training process, so that the review result is more accurate, which improves the accuracy of the published content review model in reviewing the target published content.

[0153] Based on the above embodiment, an embodiment is provided to illustrate the process of obtaining multiple review process information and review normative information of the target published content.

[0154] In an exemplary embodiment, as shown in FIG8 , obtaining multiple review process information and review normative information of target published content includes:

[0155] S801, obtain the review guidance document and review specification document for the published content review.

[0156] Among them, the audit guidance document refers to the document that guides and designs the process and steps of reviewing and publishing content, including multiple audit processes and the process information corresponding to each audit process. The audit specification document refers to the document that records the rules and regulations related to the review of published content.

[0157] In the embodiment of the present application, the audit guidance file and the audit specification file have been stored in the database when the content was reviewed in the historical time period, and can be obtained from the database. For example, the identification information of the audit guidance file and the identification information of the audit specification file can be obtained, and then the corresponding audit guidance file and audit specification file can be matched from the database based on the two identification information.

[0158] S802, extracting each audit process information according to the audit guidance document, and extracting audit normative information according to the audit specification document.

[0159] After obtaining the audit guidance document and the audit specification document, extract the information in each audit process from the audit guidance document and use it as the information of each audit process, and extract each piece of data from the audit specification document and use it as the audit normative information.

[0160] In the content review method provided in the embodiment of the present application, the review guidance document and the review specification document for the published content review are obtained, and then the review process information is extracted according to the review guidance document, and the review normative information is extracted according to the review specification document. In this method, an optional way to obtain multiple review process information and review normative information is provided. By extracting the review process information from the review guidance document and the review normative information from the review specification document, the comprehensiveness and accuracy of the information are ensured, and data support is provided for the subsequent generation of prompt words. At the same time, the accuracy and comprehensiveness of the prompt words are also improved, thereby improving the accuracy of the review results.

[0161] In addition, in an exemplary embodiment, the present application also provides a content review method, as shown in FIG9 , which may include the following steps:

[0162] S901, in response to a review request for target published content, obtaining a review guidance document and a review specification document for reviewing the published content.

[0163] S902, extracting each audit process information according to the audit guidance document, and extracting audit normative information according to the audit specification document.

[0164] S903, obtaining prompt word generation instructions for each review process.

[0165] S904, inputting each prompt word generation instruction, each audit process information and audit normative information into the prompt word generation model to obtain the audit prompt words of each audit process.

[0166] S905, performing vector conversion processing on the multimodal data in the target published content to obtain a high-dimensional vector of the multimodal data.

[0167] S906, obtaining multiple similar published contents and historical review results of each similar published content from a historical review database according to the high-dimensional vector.

[0168] S907, using each similar published content and each historical review result, the preset generative large model is trained to obtain a published content review model.

[0169] S908, obtaining standard business data of the target published content.

[0170] S909, input the standard business data, target published content and various review prompt words into the published content review model to obtain the review result.

[0171] The above-mentioned processes of S901-S909 can refer to the description of the above-mentioned method embodiment, and the implementation principles and technical effects thereof are similar and will not be repeated here.

[0172] Based on the same inventive concept, the embodiment of the present application also provides a content review system. In an exemplary embodiment, the content review system includes a review model construction module, a review prompt word generation module, and a review result generation module; the review prompt word generation module is respectively connected to the review model construction module and the review result generation module;

[0173] An audit model building module is used to obtain multiple similar published contents and historical audit results of each similar published content according to the multimodal data in the target published content, and use each similar published content and each historical audit result to perform model training to obtain a published content audit model;

[0174] An audit prompt word generation module is used to process multiple audit process information and audit normative information of target published content using a preset prompt word generation model to generate audit prompt words for multiple audit processes;

[0175] The audit result generation module is used to process the target published content and each audit prompt word using the published content audit model to obtain the audit result of the target published content.

[0176] In an embodiment of the present application, a published content review model is first constructed. Based on the multimodal data in the target published content, multiple similar published contents and historical review results of each similar published content are obtained from a historical review database, and each similar published content and each historical review result are used to perform model training to obtain a published content review model.

[0177] Secondly, audit prompt words for each audit process are generated by inputting multiple audit process information and audit normative information of the target published content into a preset prompt word generation model for processing, thereby generating audit prompt words for multiple audit processes.

[0178] Finally, the audit prompt words of each audit process are used to audit the target published content. The audit result of the target published content is obtained by inputting the target published content and each audit prompt word into the published content audit model for processing.

[0179] In the content review system provided in the embodiment of the present application, the content review system includes a review model construction module, a review prompt word generation module, and a review result generation module; the review prompt word generation module is communicated with the review model construction module and the review result generation module respectively; the review model construction module is used to obtain multiple similar published contents and historical review results of each similar published content according to the multimodal data in the target published content, and use each similar published content and each historical review result to perform model training to obtain a published content review model; the review prompt word generation module is used to use a preset prompt word generation model to process multiple review process information and review normative information of the target published content, and generate review prompt words for multiple review processes; the review result generation module is used to use the published content review model to process the target published content and each review prompt word to obtain the review result of the target published content. In this system, when the target published content is reviewed, the review is conducted based on the review prompt words of each review process, and the review prompt words of each review process are generated by inputting multiple review process information and review normative information of the target published content into the prompt word generation model, which is equivalent to extracting the review prompt words from the review process information and review normative information when reviewing the published content, and then conducting content review, which greatly reduces the work of reviewers in interpreting the review normative information and the review process, and improves the review efficiency. In addition, when generating the review prompt words, multiple review process information is also input into the prompt word generation model, which is equivalent to using the model to learn each step of the review process on the basis of the existing review process, and generate review prompt words to assist in content review. In this way, by conducting content review without modifying the original review business process, compliance with the review requirements is ensured, and the accuracy of the review results is also guaranteed.

[0180] The implementation principle, process and technical effects of each step in the above embodiment are the same as those in the above embodiment and will not be repeated. In addition, each step in each embodiment of the present application is not mandatory, and other steps may also be included between the steps. The steps in each embodiment of the present application are not limited to be performed in the described order. Some steps in each embodiment of the present application can be performed in parallel.

[0181] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0182] Based on the same inventive concept, the embodiment of the present application also provides a content audit device for implementing the content audit method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more content audit device embodiments provided below can refer to the limitations on the content audit method above, and will not be repeated here.

[0183] In an exemplary embodiment, as shown in FIG10 , a content audit device 1 is provided, comprising: an information acquisition module 10, a prompt word generation module 20 and a content audit module 30, wherein:

[0184] The information acquisition module 10 is used to obtain a plurality of audit process information and audit normative information of the target published content in response to the audit request of the target published content;

[0185] The prompt word generation module 20 is used to input each audit process information and audit normative information into a preset prompt word generation model to generate audit prompt words for multiple audit processes;

[0186] The content review module 30 is used to review the target published content according to the review prompt words of each review process to obtain the review result of the target published content.

[0187] In one embodiment, the prompt word generating module 20 is further used for:

[0188] Obtain prompt word generation instructions for each audit process; input each prompt word generation instruction, each audit process information and audit normative information into the prompt word generation model to obtain the audit prompt words for each audit process.

[0189] In one embodiment, the content review module 30 is further used to:

[0190] The multimodal data in the target published content is vectorized to obtain a high-dimensional vector of the multimodal data; the target published content is reviewed based on the high-dimensional vector and each review prompt word to obtain a review result.

[0191] In one embodiment, the content review device 1 further includes:

[0192] The data preprocessing module is used to preprocess the multimodal data to obtain preprocessed multimodal data; the preprocessing method includes at least one of data cleaning processing, word segmentation processing and syntactic analysis processing.

[0193] In one embodiment, the content review module 30 is further used to:

[0194] According to the high-dimensional vector, multiple similar published contents and the historical review results of each similar published content are obtained from the historical review database; each similar published content and each historical review result are used to train the preset generative large model to obtain a published content review model; according to each review prompt word and the published content review model, the target published content is reviewed to obtain the review result.

[0195] In one embodiment, the content review module 30 is further used to:

[0196] Obtain historical standard business data of each similar published content; input each historical standard business data, each similar published content and each audit prompt word into the generative big model to obtain the audit prediction information output by the generative big model;

[0197] According to the audit prediction information and historical audit results, the parameters of the generative big model are adjusted to obtain the published content audit model.

[0198] In one embodiment, the content review module 30 is further used to:

[0199] According to the audit prediction information, historical audit results and the preset loss function, the predicted loss value is determined; based on the predicted loss value, the model parameters of the generative large model are adjusted until the predicted loss value is less than the preset threshold, and the published content audit model is obtained.

[0200] In one embodiment, the content review module 30 is further used to:

[0201] Obtain standard business data of the target published content; input the standard business data, the target published content and various review prompts into the published content review model to obtain the review result.

[0202] In one embodiment, the prompt word generating module 20 is further used for:

[0203] Obtain the audit guidance documents and audit specification documents for the published content audit; extract the information of each audit process based on the audit guidance documents, and extract the audit normative information based on the audit specification documents.

[0204] Each module in the above-mentioned content audit device can be implemented in whole or in part by software, hardware and their combination. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the corresponding operations of each of the above modules.

[0205] In an exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be shown in FIG11. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store content audit data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a content audit method is implemented.

[0206] Those skilled in the art will understand that the structure shown in FIG. 11 is merely a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

[0207] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0208] In response to a review request for target published content, obtaining multiple review process information and review normative information of the target published content;

[0209] Input each audit process information and audit normative information into a preset prompt word generation model to generate audit prompt words for multiple audit processes;

[0210] According to the audit prompt words of each audit process, the target published content is audited to obtain the audit result of the target published content.

[0211] The implementation principles and technical effects of each step implemented by the processor in the embodiment of the present application are similar to the principles of the above-mentioned content review method and will not be repeated here.

[0212] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0213] In response to a review request for target published content, obtaining multiple review process information and review normative information of the target published content;

[0214] Input each audit process information and audit normative information into a preset prompt word generation model to generate audit prompt words for multiple audit processes;

[0215] According to the audit prompt words of each audit process, the target published content is audited to obtain the audit result of the target published content.

[0216] The implementation principles and technical effects of each step implemented when the computer program in the embodiment of the present application is executed by the processor are similar to the principles of the above-mentioned content review method and will not be repeated here.

[0217] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:

[0218] In response to a review request for target published content, obtaining multiple review process information and review normative information of the target published content;

[0219] Input each audit process information and audit normative information into a preset prompt word generation model to generate audit prompt words for multiple audit processes;

[0220] According to the audit prompt words of each audit process, the target published content is audited to obtain the audit result of the target published content.

[0221] The implementation principles and technical effects of each step implemented when the computer program in the embodiment of the present application is executed by the processor are similar to the principles of the above-mentioned content review method and will not be repeated here.

[0222] It should be noted that the personnel information (including but not limited to personnel equipment information, personnel personal information, etc.) and data (including but not limited to data for analysis, data for storage, data for display, etc.) involved in this application are all information and data authorized by the personnel or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations. In addition, in the embodiments of this application, some software, components, models and other existing solutions in the industry may be mentioned, which should be considered as exemplary, and their purpose is only to illustrate the feasibility of the implementation of the technical solution of this application, but it does not mean that the applicant has or will necessarily use the solution.

[0223] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0224] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0225] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A content review method, characterized in that: The method comprises: In response to a review request for a target published content, obtaining a plurality of review process information and review normative information of the target published content; Inputting each of the audit process information and the audit normative information into a preset prompt word generation model to generate audit prompt words for multiple audit processes; According to the audit prompt words of each audit process, the target published content is audited to obtain the audit result of the target published content.

2. The method according to claim 1, characterized in that The step of inputting each of the audit process information and the audit normative information into a preset prompt word generation model to generate audit prompt words for multiple audit processes includes: Obtaining prompt word generation instructions for each of the review processes; The prompt word generation instructions, the audit process information and the audit normative information are input into the prompt word generation model to obtain the audit prompt words of each audit process.

3. The method according to claim 1 or 2, characterized in that: The step of reviewing the target published content according to the review prompt words of each review process to obtain the review result of the target published content includes: Performing vector conversion processing on the multimodal data in the target published content to obtain a high-dimensional vector of the multimodal data; The target published content is audited according to the high-dimensional vector and each of the audit prompt words to obtain the audit result.

4. The method according to claim 3, characterized in that Before performing vector conversion processing on the multimodal data in the target published content to obtain a high-dimensional vector of the multimodal data, the method further includes: The multimodal data is preprocessed to obtain preprocessed multimodal data; the preprocessing method includes at least one of data cleaning processing, word segmentation processing and syntactic analysis processing.

5. The method according to claim 3, characterized in that: The step of reviewing the target published content according to the high-dimensional vector and each of the review prompt words to obtain the review result includes: According to the high-dimensional vector, a plurality of similar published contents and historical review results of each of the similar published contents are obtained from a historical review database; Using the similar published contents and the historical review results, the preset generative large model is trained to obtain a published content review model; According to each of the audit prompt words and the published content audit model, the target published content is audited to obtain the audit result.

6. The method according to claim 5, characterized in that The similar published contents and the historical audit results are used to train the preset generative large model to obtain a published content audit model, including: Obtaining historical standard business data of each of the similar published contents; Inputting each of the historical standard business data, each of the similar published contents and each of the audit prompt words into the generative big model to obtain the audit prediction information output by the generative big model; According to the audit prediction information and the historical audit results, the parameters of the generative big model are adjusted to obtain the published content audit model.

7. The method according to claim 6, characterized in that The step of adjusting the parameters of the generative large model according to the audit prediction information and each of the historical audit results to obtain the published content audit model includes: Determine a predicted loss value according to the audit prediction information, each of the historical audit results and a preset loss function; Based on the predicted loss value, the model parameters of the generative large model are adjusted until the predicted loss value is less than a preset threshold, thereby obtaining the published content review model.

8. The method according to claim 5, characterized in that The step of reviewing the target published content according to each of the review prompt words and the published content review model to obtain the review result includes: Obtaining standard business data of the target published content; The standard business data, the target published content and each of the review prompt words are input into the published content review model to obtain the review result.

9. The method according to claim 1 or 2, characterized in that: The obtaining of multiple review process information and review normative information of the target published content includes: Obtain the review guidance and review specification documents for published content review; According to the audit guidance document, the audit process information is extracted, and according to the audit specification document, the audit normative information is extracted.

10. A content review system, characterized in that: The content audit system includes an audit model construction module, an audit prompt word generation module, and an audit result generation module; the audit prompt word generation module is respectively connected to the audit model construction module and the audit result generation module in communication; The audit model building module is used to obtain multiple similar published contents and historical audit results of each similar published content according to the multimodal data in the target published content, and use each similar published content and each historical audit result to perform model training to obtain a published content audit model; The review prompt word generation module is used to process multiple review process information and review normative information of the target published content using a preset prompt word generation model to generate review prompt words for multiple review processes; The audit result generating module is used to process the target published content and each of the audit prompt words using the published content audit model to obtain the audit result of the target published content.

11. A content review device, characterized in that: The device comprises: An information acquisition module, configured to respond to a review request for a target published content and acquire a plurality of review process information and review normative information of the target published content; A prompt word generation module, used for inputting each of the audit process information and the audit normative information into a preset prompt word generation model to generate audit prompt words for multiple audit processes; The content review module is used to review the target published content according to the review prompt words of each review process to obtain the review result of the target published content.

12. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 9 are implemented.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 9 are implemented.

14. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 9 are implemented.

Citation Information

Cited By

  • Authorization plan processing method and device based on airworthiness authentication platform

    CN121599440A

  • An automated rule review system for fine-tuning large models

    CN123092238A