Risk auditing method and device for data work order, storage medium and terminal

Abstract

The embodiment of the invention discloses a data work order risk auditing method and device, a storage medium and a terminal. Extracting feature data of the to-be-audited work order; inputting the feature data into a risk auditing large model, and determining a risk auditing result of the to-be-audited work order output by the risk auditing large model based on the feature data; if the risk checking result is that the risk does not exist, confirming that the to-be-checked work order passes the checking; and if the risk checking result is that the risk exists, marking the work order to be checked as a risk work order. The risk auditing large model can intelligently identify and distinguish risk-free work orders and potential risk work orders by using a deep learning algorithm and mass data analysis capability. The application of the risk auditing large model reduces the manual intervention link in the auditing process, and significantly improves the efficiency and accuracy of work order auditing.

Description

Technical Field

[0001] The embodiments of this specification relate to the field of artificial intelligence technology, and in particular, to a risk review method, device, storage medium, and terminal for a data work order. Background Art

[0002] With the continuous enrichment and expansion of platform scenarios, the number of various marketing advertisements has increased, from traditional graphic advertisements to today's popular short videos, live streaming and other diversified forms, which makes the audit work face greater challenges. The audit team not only needs to deal with the surge in quantity, but also needs to confirm whether the preferential information, product descriptions, etc. in the advertisements are accurate to ensure the authenticity, legality and compliance of each advertisement content. At the same time, the timeliness requirements on the audit side have not decreased but increased. Therefore, how to effectively improve the efficiency of auditing has become an important issue that needs to be solved urgently. Summary of the invention

[0003] The embodiments of this specification provide a risk review method, device, storage medium and terminal for a data work order, which can solve the technical problems of high cost and low efficiency of manual review in related technologies.

[0004] In a first aspect, an embodiment of the present specification provides a risk review method for a data work order, the method comprising:

[0005] Extract feature data from the work orders to be reviewed;

[0006] Input the characteristic data into the risk review model, and determine the risk review results of the work orders to be reviewed output by the risk review model based on the characteristic data;

[0007] If the risk review result is that there is no risk, the work order to be reviewed is confirmed to have passed the review;

[0008] If the risk review result is that there is a risk, the work order to be reviewed will be marked as a risky work order.

[0009] In a possible implementation, the above-mentioned determination of the risk review result of the work order to be reviewed by the risk review big model based on the characteristic data includes: determining the risk review result of the work order to be reviewed output by the risk review big model after analyzing the characteristic data according to at least one prompt word; the prompt word is obtained after configuration based on at least one type of risk information, and the prompt word is used to specify the recognition purpose and output rules of the risk review big model.

[0010] In a possible implementation, the types of risk information include at least one of content authenticity risk, content and product relevance risk, content security risk, and intellectual property risk; when there are at least two risk types, each prompt word is configured based on a different risk type.

[0011] In a possible implementation, when there are at least two prompt words, the risk review model is configured to divide the at least two prompt words into multiple prompt word batches, and call each prompt word batch in sequence; wherein each prompt word batch includes at least one prompt word.

[0012] In a possible implementation, the above-mentioned extraction of feature data for the work order to be reviewed includes: when it is recognized that the work order to be reviewed includes text, extracting text feature data of the text; when it is recognized that the work order to be reviewed includes a picture, using the picture as the picture feature data of the work order to be reviewed, and performing picture text recognition on the picture to obtain the text feature data of the picture; the method of picture text recognition includes at least one of an optical character recognition method and a visual language model recognition method.

[0013] In a possible implementation, the above-mentioned feature data is text feature data, and the above-mentioned determination of the risk review result of the work order to be reviewed based on the feature data output by the risk review big model includes: controlling the risk review big model to call the text analysis model to analyze the text feature data, and outputting the risk review result of the work order to be reviewed.

[0014] In a possible implementation, the feature data is image feature data, and the risk review result of the work order to be reviewed output by the risk review big model based on the feature data is determined, including: controlling the risk review big model to call the visual language model to analyze the image feature data, and outputting the risk review result of the work order to be reviewed.

[0015] In a possible implementation, after extracting the feature data from the work order to be reviewed, the process also includes: determining whether the work order to be reviewed needs to be input into the risk review model for review; if so, executing the step of inputting the feature data into the risk review model; if not, transferring the work order to be reviewed to the manual review node.

[0016] In a possible implementation, the above-mentioned determination of whether the work order to be reviewed needs to be input into the risk review big model for review includes: determining whether the type of characteristic data in the work order to be reviewed meets the calling conditions of the risk review big model; if not, determining that the work order to be reviewed does not need to be input into the risk review big model for review.

[0017] In one possible implementation, the above-mentioned judgment of whether the work order to be reviewed needs to be input into the risk review big model for review includes: pre-filtering the work order to be reviewed, the pre-filtering includes at least one processing method of keyword matching and historical data deduplication; if the work order to be reviewed passes the pre-filtering, it is determined that the work order to be reviewed needs to be input into the risk review big model for review; if the work order to be reviewed does not pass the pre-filtering, it is determined that the work order to be reviewed does not need to be input into the risk review big model for review, and the work order to be reviewed is transferred to the re-inspection node.

[0018] In a possible implementation, after marking the work order to be reviewed as a risky work order, the process further includes: transferring the risky work order to a re-inspection node, and determining the re-inspection result of the risky work order in the re-inspection node; if the re-inspection result is that there is no risk, confirming that the work order to be reviewed has passed the review; if the re-inspection result is that there is a risk, intercepting and rejecting the work order to be reviewed.

[0019] In a possible implementation, the review node includes at least one of a manual review node and a neural network model review node.

[0020] In a possible implementation, the method further includes: performing quality inspection on the risk audit big model according to a preset period, and updating the configuration of the risk audit big model according to the quality inspection results; the quality inspection includes at least one of the audit result quality inspection and the model architecture quality inspection.

[0021] In a possible implementation, the above-mentioned work order to be reviewed is a marketing advertising work order.

[0022] In a second aspect, an embodiment of the present specification provides a risk review device for a data work order, the device comprising:

[0023] Data acquisition module, used to extract feature data from the work orders to be reviewed;

[0024] The model review module is used to input the characteristic data into the risk review model and determine the risk review results of the work orders to be reviewed output by the risk review model based on the characteristic data;

[0025] The review pass module is used to confirm that the pending work order has passed the review if the risk review result is that there is no risk;

[0026] The risk marking module is used to mark the work order to be reviewed as a risk work order if the risk review result is that there is a risk.

[0027] In a third aspect, an embodiment of the present specification provides a computer program product comprising instructions, which, when executed on a computer or a processor, enables the computer or the processor to execute the steps of the above method.

[0028] In a fourth aspect, an embodiment of the present specification provides a computer storage medium, wherein the computer storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded by a processor and executing the steps of the above method.

[0029] In a fifth aspect, an embodiment of the present specification provides a terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is suitable for being loaded by the processor and executing the steps of the above method.

[0030] The beneficial effects brought by the technical solutions provided by some embodiments of this specification include at least:

[0031] The embodiment of this specification provides a risk audit method for a data work order, extracting feature data from the work order to be audited; inputting the feature data into a risk audit big model, and determining the risk audit result of the work order to be audited output by the risk audit big model based on the feature data; if the risk audit result is that there is no risk, then confirm that the work order to be audited has passed the audit; if the risk audit result is that there is a risk, then mark the work order to be audited as a risky work order. By introducing the risk audit big model into the audit process, automatic audit of work orders can be achieved. The risk audit big model can use deep learning algorithms and massive data analysis capabilities to intelligently identify and distinguish between risk-free work orders and potential risk work orders. For those work orders that are confirmed to be risk-free after model evaluation, the system will immediately pass them to ensure the smooth progress of normal business processes; and for the work orders that are identified to be at risk, the big model can accurately mark them for subsequent processing. The application of the risk audit big model reduces the manual intervention in the audit process, significantly improves the efficiency and accuracy of work order audits, reduces the operating costs of enterprises, and enhances risk prevention and control capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this specification. For those skilled in the art, other drawings can be obtained based on these drawings without paying any creative work.

[0033] Figure 1 An exemplary system architecture diagram of a risk review method for a data work order provided in an embodiment of this specification;

[0034] Figure 2 A schematic diagram of a process flow of a risk review method for a data work order provided in an embodiment of this specification;

[0035] Figure 3 A schematic diagram of a process flow of a risk review method for a data work order provided in an embodiment of this specification;

[0036] Figure 4 A schematic diagram of a process flow of a risk review method for a data work order provided in an embodiment of this specification;

[0037] Figure 5 A logical framework diagram of a flow control strategy provided in an embodiment of this specification;

[0038] Figure 6 A process framework diagram of a risk audit method provided in an embodiment of this specification;

[0039] Figure 7 A schematic diagram of a process flow of a risk review method for a data work order provided in an embodiment of this specification;

[0040] Figure 8 A structural block diagram of a risk review device for a data work order provided in an embodiment of this specification;

[0041] Fig. 9 A schematic diagram of the structure of a terminal provided in an embodiment of this specification. DETAILED DESCRIPTION

[0042] In order to make the features and advantages of the embodiments of this specification more obvious and easy to understand, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this specification.

[0043] When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this specification. Instead, they are merely examples of devices and methods consistent with some aspects of this specification as detailed in the appended claims. And in the description of the embodiments of this specification, unless otherwise indicated, " / " means or, for example, A / B can mean A or B: "and / or" in the text is only a description of the association relationship of associated objects, indicating that there can be three relationships, such as A and / or B, which can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of this specification, "multiple" refers to two or more than two.

[0044] In the following, the terms "first" and "second" are used for descriptive purposes only and are not to be understood as suggesting or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features.

[0045] As the user groups and merchant groups on the platform become increasingly large, the number and types of marketing advertisements have also shown a blowout growth. This means that auditors not only need to deal with the surge in the number of advertisements, but also need to ensure the authenticity, legality and compliance of the content of each advertisement. In addition to the complex and time-consuming steps involved in the audit process, marketing activities, as a key means to drive platform traffic and user activity, also need to be planned and executed in a rhythm that responds quickly to market changes, which leads to extremely high timeliness requirements on the audit side. However, in actual scenarios, on the premise of ensuring the quality of the audit, the timeliness of the audit side is generally between 0.5 and 2 hours. Such a processing speed is obviously difficult to meet the requirements of rapid response in the face of intensive releases of marketing activities and frequent urgent online demands. It may even be easy to cause blockages in the audit process and delays in the launch of marketing activities, affecting the overall operational efficiency and user experience of the platform.

[0046] Therefore, the embodiments of this specification provide a risk review method for a data work order to solve the above-mentioned technical problems of high cost and low efficiency of manual review.

[0047] See also Figure 1 , Figure 1 An exemplary system architecture diagram of a risk review method for a data work order provided in an embodiment of this specification.

[0048] like Figure 1 As shown, the system architecture may include a terminal 101, a network 102, and a server 103. The network 102 is used to provide a medium for a communication link between the terminal 101 and the server 103. The network 102 may include various types of wired communication links or wireless communication links, for example, the wired communication link includes an optical fiber, a twisted pair, or a coaxial cable, and the wireless communication link includes a Bluetooth communication link, a Wireless-Fidelity (Wi-Fi) communication link, or a microwave communication link.

[0049] The terminal 101 can interact with the server 103 through the network 102 to receive messages from the server 103 or send messages to the server 103, or the terminal 101 can interact with the server 103 through the network 102 to receive messages or data sent by other users to the server 103. The terminal 101 can be hardware or software. When the terminal 101 is hardware, it can be various electronic devices, including but not limited to smart phones, tablet computers, laptop portable computers and desktop computers. When the terminal 101 is software, it can be installed in the electronic devices listed above, which can be implemented as multiple software or software modules (for example: used to provide distributed services), or it can be implemented as a single software or software module, which is not specifically limited here.

[0050] In the embodiment of the present specification, terminal 101 first extracts feature data from the work order to be reviewed; then terminal 101 inputs the feature data into the risk review big model to determine the risk review result of the work order to be reviewed output by the risk review big model based on the feature data; if the risk review result is that there is no risk, terminal 101 confirms that the work order to be reviewed has passed the review; if the risk review result is that there is a risk, terminal 101 marks the work order to be reviewed as a risky work order.

[0051] The server 103 may be a business server that provides various services. It should be noted that the server 103 may be hardware or software. When the server 103 is hardware, it may be implemented as a distributed server cluster consisting of multiple servers, or it may be implemented as a single server. When the server 103 is software, it may be implemented as multiple software or software modules (for example, for providing distributed services), or it may be implemented as a single software or software module, which is not specifically limited here.

[0052] Alternatively, the system architecture may not include the server 103. In other words, the server 103 may be an optional device in the embodiments of this specification, that is, the method provided in the embodiments of this specification may be applied to a system structure that only includes the terminal 101, and the embodiments of this specification do not limit this.

[0053] It should be understood that Figure 1 The number of terminals, networks and servers in the figure is only for illustration and any number of terminals, networks and servers may be used according to implementation requirements.

[0054] See also Figure 2 , Figure 2A flowchart of a risk review method for a data work order provided in an embodiment of this specification. The execution subject of the embodiment of this specification can be a terminal that executes the risk review of the data work order, or a processor in the terminal that executes the risk review method for the data work order, or a risk review service for the data work order in the terminal that executes the risk review method for the data work order. For the convenience of description, the specific execution process of the risk review method for the data work order is introduced below by taking the execution subject being the processor in the terminal as an example.

[0055] like Figure 2 As shown, the risk review method for data work orders can at least include:

[0056] S202. Extract feature data from the work order to be reviewed.

[0057] Optionally, generally after users upload content, each piece of content will be in the form of a work order waiting for review in the system. After the reviewers conduct a multi-faceted review of these pending work orders to ensure that the content is legal and accurate, the content will be opened to the platform. For example, on an advertising platform, merchants need to upload their own advertisements to the platform to achieve their own publicity and marketing purposes. After the merchant uploads the advertising content, a corresponding marketing advertising work order will be generated. These work orders will be subject to risk review by the platform as pending work orders, thereby ensuring that risk-free advertising content is put online to the user end.

[0058] Optionally, for a large number of work orders to be reviewed, the embodiments of this specification take into account the performance advantages of large models and introduce large risk review models to improve the efficiency and timeliness issues faced when manually reviewing work orders. Specifically, the large risk review model can use deep learning algorithms and massive data analysis capabilities to intelligently identify and distinguish between risk-free work orders and potential risk work orders. In addition to the obvious improvement over manual review, the large risk review model also has advantages in many aspects compared to the traditional review system based on small-volume models. First, in terms of training costs, the large model benefits from its powerful data processing capabilities and algorithm optimization. While ensuring high accuracy, it can significantly reduce the reliance on a large amount of labeled data, reducing the cost investment in tedious links such as data collection, cleaning, and labeling. Secondly, traditional small models are often limited by the quality of the corpus in a specific field or scenario, while the large risk review model can maintain stable performance across fields and scenarios through advanced transfer learning and generalization capabilities.

[0059] Furthermore, when processing the work orders to be reviewed through the risk review model, the first thing to do is to extract the feature data. By accurately capturing and organizing the content of the work orders to be reviewed, it is possible to identify the comprehensive and detailed feature data of the work orders to be reviewed, providing a solid foundation for subsequent risk assessment. The types of feature data include but are not limited to text and images, which specifically include information on multiple dimensions such as the submission time of the work orders to be reviewed, the identity of the submitter, the category to which they belong, the product content description, the amount involved, and marketing materials.

[0060] S204: Input the characteristic data into the risk review big model, and determine the risk review result of the work order to be reviewed output by the risk review big model based on the characteristic data.

[0061] Optionally, after extracting the feature data, the feature data set is input into a pre-trained risk review model. The risk review model can comprehensively analyze the feature data based on its own data processing and analysis capabilities, and finally give the risk review results of the work order to be reviewed. Compared with the original 0.5-2 hour response time for manual review, the risk review model can improve the analysis time of each work order to seconds. The application of the risk review model reduces the manual intervention in the review process, significantly improves the efficiency and accuracy of work order review, reduces the company's operating costs, and enhances risk prevention and control capabilities.

[0062] In a possible implementation, the risk audit big model can be trained based on the basic multimodal big model, and the basic multimodal big model can output prediction results for the prediction object based on various types of features of the prediction object. Therefore, by constructing an initial risk audit big model for the risk audit scenario based on the basic multimodal big model, the initial risk audit big model can also obtain the ability to "output prediction results for the prediction object based on various types of features of the prediction object". Furthermore, after constructing the initial risk audit big model for the risk audit scenario, the initial risk audit big model is trained in a targeted manner for the application scenario, and the parameters of the initial risk audit big model are continuously adjusted during the training process until the initial risk audit big model converges to obtain a risk audit big model that can be put into online application.

[0063] S206. If the risk review result is that there is no risk, confirm that the work order to be reviewed has passed the review.

[0064] Optionally, the next step of judgment and processing can be carried out according to the output results of the risk review model. If the result shows that there is no risk in the work order to be reviewed, it can be confirmed that the work order has passed the review process and can directly enter the next stage of operation, such as uploading the user end or notifying the relevant parties.

[0065] S208. If the risk review result is that there is a risk, the work order to be reviewed will be marked as a risky work order.

[0066] However, if the risk review results show that the work order to be reviewed is risky, the work order will be marked as a risky work order. System administrators or risk management departments can conduct more in-depth analysis and judgment based on these marks, and take necessary risk mitigation measures. For risky work orders, additional review processes may be initiated, including manual review, expert review, or request for additional materials to ensure that all potential risks are properly handled and the overall process is safe and stable.

[0067] In an embodiment of the present specification, a risk audit method for a data work order is provided, wherein feature data is extracted from the work order to be audited; the feature data is input into a risk audit big model, and the risk audit result of the work order to be audited output by the risk audit big model based on the feature data is determined; if the risk audit result is that there is no risk, the work order to be audited is confirmed to have passed the audit; if the risk audit result is that there is a risk, the work order to be audited is marked as a risky work order. By introducing a risk audit big model into the audit process, automatic audit of work orders can be achieved. The risk audit big model can use deep learning algorithms and massive data analysis capabilities to intelligently identify and distinguish between risk-free work orders and potential risk work orders. For those work orders that are confirmed to be risk-free after model evaluation, the system will immediately pass them to ensure the smooth progress of normal business processes; and for the work orders that are identified to be at risk, the big model can accurately mark them for subsequent processing. The application of the risk audit big model reduces the manual intervention in the audit process, significantly improves the efficiency and accuracy of work order audits, reduces the operating costs of enterprises, and enhances risk prevention and control capabilities.

[0068] In step S204, the feature data extracted from the work order to be reviewed is input into the risk review model, and the risk review model is used to analyze whether there are risks in the feature data and what risks exist, and then the risk review results of the work order to be reviewed are output, which effectively improves the efficiency of the work order review. In the big model, prompts play a vital role, and they can significantly affect the output quality and effect of the model.

[0069] Then in a risk review method for a data work order provided in some embodiments of the present specification, step S204 specifically includes: determining the risk review result of the work order to be reviewed outputted after the risk review big model analyzes the characteristic data according to at least one prompt word.

[0070] Specifically, in order to allow the big model to perform a series of audit operations more accurately, prompt words can be configured in advance based on the risk information that needs to be identified. In the configured prompt words, the identification purpose and output rules of the risk audit big model are specified to help the big model clarify the task type and the operations that need to be performed. For example, when dealing with intellectual property risks, the risk trigger conditions and risk assessment rules corresponding to the intellectual property risks can be added to the corresponding prompt words, and the prompt words can be configured as "perform intellectual property risk identification operations on the current content, and return corresponding results based on whether there are intellectual property risks", so that the big model can more easily understand the current operations that need to be performed, the purpose of the operations, and the rules that need to be followed when returning. By providing more specific prompt words, the big model can more easily generate outputs that meet expectations.

[0071] Furthermore, in actual scenarios, there are often multiple risks that can easily lead to adverse consequences, such as content authenticity risks, content and product relevance risks, content security risks, intellectual property risks, etc. Therefore, each work order to be reviewed needs to face multiple risk reviews and investigations. In order to deal with multiple risks, corresponding prompt words can be configured for the risk information of various risks, that is, when there are at least two risk types, each prompt word is configured based on different risk types. By changing the prompt words, the large model can be guided to make corresponding judgment operations when reviewing different risks, which increases the flexibility of the large model during review. Appropriate prompt word configuration is also conducive to the stability of the large model when conducting multiple risk analyses.

[0072] Based on this, when the risk audit model audits the feature data, it specifically analyzes the feature data according to at least one prompt word, and then outputs the risk audit results of the work order to be audited. When the risk audit model calls too many prompt words at one time, the risk audit model will process too much information at one time, resulting in timeouts, freezes, or poor output effects. Therefore, when there are at least two prompt words, the risk audit model can be set to divide all prompt words into multiple prompt word batches, each batch includes at least one prompt word, and then each prompt word is called in turn according to the batch during each audit.

[0073] For example, in the audit scenario of marketing advertisements, 15 prompt words are configured for 15 risks. Then the risk audit model divides the 15 prompt words into 3 batches, each batch includes 5 prompt words. The risk audit model first audits each work order to be audited according to the 5 prompt words in the first batch, and outputs the risk results of the work order in these 5 aspects; then audits according to the 5 prompt words in the second batch, and so on, until all batches of prompt words are called, and all risk results are uniformly output as risk audit results of the work order to be audited. This arrangement of calling multiple prompt words in batches is conducive to maintaining the stability and output efficiency of the large model.

[0074] In the embodiment of this specification, a risk audit method for data work orders is provided, prompt words are configured in advance based on multiple risk information, and when there are multiple prompt words, the risk audit big model is set to call the prompt words in batches, and the feature data is analyzed according to the prompt words. In this way, by configuring prompt words and reasonably arranging the use of prompt words by the big model, it can not only help the big model better understand the task and improve the output quality, but also increase diversity and reduce misleading, and finally achieve the effect of optimizing the performance of the big model and user experience.

[0075] There may be many different forms of information in the work order to be reviewed. For different types of information, the risk review model needs to use targeted information analysis methods, risk identification algorithms, etc. Based on this, this specification takes text data and image data as examples to describe the specific process of the big model analyzing different types of data information. Please refer to Figure 3 , Figure 3 A flowchart of a risk review method for a data work order provided in an embodiment of this specification.

[0076] like Figure 3 As shown, step S202 in the risk review method for data work orders may at least include:

[0077] S302: When it is identified that the work order to be reviewed includes text, extract text feature data of the text.

[0078] Optionally, in the process of processing the work order to be reviewed, the system can automatically identify and process multiple data types contained in the work order. Specifically, when it is identified that the work order to be reviewed contains text content, the text feature extraction module will be immediately started to extract the text feature data of the text content. Text feature data may include keywords, phrases, sentence structure, sentiment tendency, topic classification and other aspects, which together constitute a multi-dimensional representation of the text content, providing a rich information basis for subsequent risk review.

[0079] S304. When it is identified that the work order to be reviewed includes a picture, the picture is used as the picture feature data of the work order to be reviewed, and the picture text recognition is performed on the picture to obtain the text feature data of the picture; the picture text recognition method includes at least one of an optical character recognition method and a visual language model recognition method.

[0080] Optionally, correspondingly, when it is identified that the work order to be reviewed contains a picture, considering that the pixel information of the picture itself and the text information in the picture are both important features of the picture, the picture itself and the text in the picture need to be extracted and used.

[0081] Specifically, the image itself will be saved and processed as the image feature data of the work order to be reviewed. As an intuitive and vivid information carrier, the image can often provide more details and background information about the content of the work order. In addition to using the image as direct feature data, the image needs to be further processed for image text recognition. The purpose of this step is to extract the text information contained in the image. This text information may exist in the image in the form of text, numbers, symbols, etc., and is an important part of the image content. There are many ways to perform image text recognition, and in the embodiments of this specification, at least one of an optical character recognition method and a visual language model recognition method is included.

[0082] Among them, when extracting text through optical character recognition (OCR), the system mainly uses optical methods to convert the text in the paper document into a black and white dot matrix image file, and converts the text in the image into text format through recognition software, and then further processes the text to obtain the text data of the work order to be reviewed. When extracting text through visual language model recognition, the system can call visual language models such as the Qwen-VL Max model to accurately describe and recognize image information. The performance of the model can support the recognition of high-definition resolution images and extreme aspect ratio images above one million pixels, can fully and accurately reproduce dense text, and can also extract information from tables and documents. After converting the text information in the image into text data through image text recognition technology, it is helpful to further enrich the text feature data of the work order to be reviewed.

[0083] In a possible embodiment, when extracting text feature data of text and images, the system will comprehensively consider multiple factors to ensure the accuracy and completeness of the data. For example, for text feature data, the system will perform operations such as word frequency statistics, keyword extraction, and sentiment analysis; for image feature data, it will perform operations such as image recognition, color analysis, and texture analysis. At the same time, the system will also clean and deduplicate the extracted feature data to eliminate redundant and erroneous information and improve the quality and availability of the data.

[0084] S306: Input the text feature data into the risk review big model, control the risk review big model to call the text analysis model to analyze the text feature data, and output the risk review result of the work order to be reviewed.

[0085] Optionally, after the text feature data is input into the risk audit model, the control risk audit model calls a text analysis model specifically used for analyzing text to analyze the text feature data, and outputs the text risk audit results of the work order to be audited. In a preferred embodiment, the text analysis model can be specifically a Qwen-Max model. Qwen-Max is based on the Transformer architecture and can better understand the context and output the target results through multiple algorithms such as deep learning and natural language processing (NLP). Therefore, using the Qwen-Max model for text processing can obtain stable and reliable text audit results.

[0086] S308. Input the image feature data into the risk review big model, control the risk review big model to call the visual language model to analyze the image feature data, and output the risk review result of the work order to be reviewed.

[0087] Optionally, after the image feature data is input into the risk audit model, the control risk audit model calls the image analysis model specifically used to analyze the image feature data, and outputs the image risk audit results of the work order to be audited. In another preferred embodiment, the risk analysis model can be specifically the Qwen-VL Max model. Qwen-VLMax is a large-scale visual language model built on Qwen-7B and ViT-G, using a three-stage training method, including weakly supervised pre-training, multi-task pre-training and supervised fine-tuning, which improves the model performance. Qwen-VL Max performs well in visual reasoning, can process high-definition images at the megapixel level, and understand images of various extreme aspect ratios. In addition, it also supports the input of images, texts and detection frames, and can output texts and detection frames to realize the processing and understanding of multimodal information. Therefore, using the Qwen-VL Max model for image processing can obtain accurate image recognition results and audit results.

[0088] S310. If the risk review result is that there is no risk, then confirm that the work order to be reviewed has passed the review; if the risk review result is that there is a risk, then mark the work order to be reviewed as a risky work order.

[0089] Regarding step S310, please refer to the detailed description in steps S206-S208, which will not be repeated here.

[0090] In the embodiment of this specification, a risk audit method for data work orders is provided. When it is identified that the work order to be audited contains text and pictures, the text feature data of the text and the picture feature data of the picture are extracted respectively, and the picture is processed by OCR to obtain the text feature data of the picture. This process not only improves the data quality of the work order to be audited, but also provides more abundant information support for subsequent audit work, which is conducive to the accuracy and comprehensiveness of the subsequent risk audit stage.

[0091] There are various forms of content to be reviewed on the platform, such as text, pictures, videos, audio, and tables. The content formats that the risk review big model can handle are usually relatively fixed, and for some work orders that are obviously illegal (such as high duplication with historical work orders, and hitting sensitive keywords), the big model has computing power overflow for the identification of such risks. Such risk reviews conducted through the big model will obviously occupy the audit computing power of the cases that really need in-depth review. Therefore, before sending the work orders to be reviewed into the risk review big model, the flow of the work orders to be reviewed can be controlled first. Through certain flow strategies, the work orders that can be reviewed by the big model and really need to be reviewed by the big model will be transferred to the big model review link, while the work orders that cannot be reviewed by the big model or do not need to be reviewed by the big model will be transferred to other links to avoid unnecessary waste of the big model computing power.

[0092] Please refer to Figure 4 , Figure 4 The following is a flow chart of a risk audit method for a data work order provided in an embodiment of this specification. Figure 4 As shown, in the risk review method for data work orders, after step S202 and before step S204, at least the following may be included:

[0093] S402: Determine whether the work order to be reviewed needs to be input into the risk review model for review.

[0094] First, please refer to Figure 5 , Figure 5 This is a logical framework diagram of a flow control strategy provided in the embodiment of this specification. Figure 5 As shown, the strategy for flow control of work orders to be reviewed needs to be pre-configured. A variety of flow control screening conditions are configured in the work order flow control strategy, so that when the system calls the strategy, it can find the review link to which each work order to be reviewed belongs according to the link logic configured in the strategy, and then implement the risk review of the work order according to the corresponding review link.

[0095] Optionally, in practical applications, such as Figure 6In the process framework diagram of a risk audit method shown in the figure, when a work order to be audited appears, the system needs to judge whether the work order to be audited needs to be input into the risk audit model for audit according to the pre-set flow control strategy, and send the work order to be audited to its corresponding audit channel according to the judgment result. Figure 6 It can be seen that in addition to the risk audit big model that can be audited in the audit system, a bottom-line audit channel is also retained to make up for the situation where the big model cannot be audited, reducing the possibility of risk leakage. Among them, the bottom-line audit channel may specifically include simple machine audit mechanisms (such as keyword matching), manual audits, and other non-big model audit methods. These audit methods can serve as a supplement and extension of the big model audit method, which is conducive to optimizing the overall audit effect of the system.

[0096] In a possible embodiment, if the feature data does not meet the calling conditions of the model, even if the work order is forcibly input into the model for review, accurate and reliable review results may not be obtained. On the contrary, doing so may also waste computing resources, reduce review efficiency, and even introduce additional errors and risks. Therefore, when performing flow control, it is possible to judge whether it is suitable for the large model risk review method based on the feature data of the work order to be reviewed, that is, to judge whether the type of feature data in the work order to be reviewed meets the calling conditions of the risk review large model. In this judgment process, the system will screen and verify the feature data according to the specific requirements of the risk review large model. If the feature data type in the work order to be reviewed does not match the calling conditions of the risk review large model, such as data type error, data missing or data format does not meet the requirements, then the system will immediately make a judgment to determine that the work order does not need to be input into the risk review large model for review.

[0097] In another possible embodiment, for some work orders with obvious security risks that can be identified by simple machine review, it is unnecessary to consume the computing resources of the large model. In this case, a detailed and efficient pre-filtering process can be performed before entering the large model. This process aims to filter out the cases that really need in-depth review from many work orders to improve the overall review efficiency and accuracy. The pre-filtering process includes multiple strategies, including at least two methods: keyword matching and historical data deduplication.

[0098] Specifically, keyword matching is a highly automated technology that quickly scans the content of the work order to be reviewed based on a series of preset keywords or phrases. These keywords are usually associated with known problem types, risk points or sensitive information. If the work order contains content that matches these keywords, the system will directly mark the work order as a risky work order, and there is no need to send it to the big model for review. On the other hand, historical data deduplication uses advanced algorithms and database technology to compare the current work order to be reviewed with the work orders that have been processed in the past, and identify and exclude those with similar or completely duplicate content. This step can effectively avoid repeated and invalid audits, saving a lot of time and resources.

[0099] After the pre-filtering process, if the pending work order passes the pre-filtering process, it means that it has not triggered obvious risks, and then it needs to be input into the risk review model for a more detailed review. On the contrary, if the pending work order fails to pass the pre-filtering process, it means that there are obvious risks in it, and there is no need to conduct a detailed review through the risk review model, but the work order can be transferred to the re-inspection node. The re-inspection node is a link composed of manual or higher-level automated systems, which is responsible for supplementary review of pending work orders that are found to have risks in the filtering link. Through such a pre-filtering processing mechanism, not only the audit efficiency is greatly improved, but also the reasonable allocation of audit resources is ensured, so that the pending work orders can receive timely and effective attention and processing.

[0100] S404: If necessary, execute the step of inputting the characteristic data into the risk review model.

[0101] Please continue reading Figure 6 ,like Figure 6 As shown, if it is found during flow control that the work order to be reviewed needs to be reviewed by a large model, then the characteristic data can be input into the risk review large model. The large model calls at least one pre-configured prompt word according to the preset prompt word calling method, and then outputs the risk review result of the work order to be reviewed.

[0102] Optionally, a fallback strategy is also provided for the audit mechanism of large models. During the audit process of large models, if any abnormal situations such as audit failure or timeout are encountered, retries will be performed. If the number of retries exceeds the preset threshold (for example, 3 times, 5 times, etc.), the large model audit will be directly abandoned and directly transferred to the fallback audit channel, and other audit methods will be used for audit, thereby compensating for the possible instability of the large model and reducing risk leakage.

[0103] S406: If not necessary, the work order to be reviewed will be transferred to the manual review node.

[0104] Please continue reading Figure 6 ,like Figure 6 As shown, if it is not necessary, it means that the work order to be reviewed cannot be reviewed by a large model or it is unnecessary to review by a large model. In this case, such a work order will be directly transferred to the manual review node, and risk leakage prevention will be carried out by manual review.

[0105] In an embodiment of the present specification, a risk review method for data work orders is provided, which controls the flow of all work orders to be reviewed based on a flow control strategy, supports inputting large models into large models, and does not need to call work orders of large models to directly flow to manual review nodes. By flexibly configuring the flow control strategy and combining various types of review methods, it can ensure that each work order to be reviewed can be properly handled while keeping the entire review process smooth and efficient.

[0106] For work orders that are marked as risky during the machine review stage (risk review model, pre-filtering process), they can be further reviewed by manual or other automated review mechanisms. The review node is responsible for conducting a second check on work orders that may be missed or misjudged during the filtering process to ensure the accuracy of the final review results. Figure 7 , Figure 7 A flowchart of a risk review method for a data work order provided in an embodiment of this specification.

[0107] like Figure 7 As shown, the risk review method of the data work order may include at least the following after step S208:

[0108] S702: Transfer the risk work order to the re-inspection node, and determine the re-inspection result of the risk work order in the re-inspection node.

[0109] Optionally, see Figure 6 ,like Figure 6 As shown, when the work order to be reviewed circulates in the system, if it is identified as a risky work order, it will be transferred to a dedicated re-inspection node. The re-inspection node includes at least one of a manual review node and a neural network model review node, where the neural network model is pre-trained and converged based on the sample risk work order data. The review method in the re-inspection node needs to be more stable because it ensures that all potential risks can receive a detailed and professional second assessment. At the re-inspection node, the risk work order is analyzed in depth, and the re-inspection result of the risk work order is determined based on the established risk assessment standards and procedures and after comprehensive consideration of various factors.

[0110] S704. If the re-inspection result is that there is no risk, confirm that the work order to be reviewed has passed the review.

[0111] Optionally, there may be two kinds of re-inspection results: one is that there is no risk, and the other is that there is a risk. For work orders that are determined to have no risk after re-inspection, it means that the risk points that were previously identified are considered to not pose an actual threat or have been effectively controlled after re-examination. Therefore, the system or manual auditor will formally confirm that these pending work orders have passed the review process and allow them to continue to the next stage of processing or execution to ensure the smooth progress of the transaction process.

[0112] S706. If the re-inspection result shows that there is a risk, the work order to be reviewed will be intercepted and rejected.

[0113] Optionally, if the re-examination results show that the risk does exist, it means that the issues involved in the work order have not been properly resolved or need further review. At this time, in order to maintain the security and compliance of the transaction process, the system will automatically or based on the auditor's judgment, intercept the work order to be reviewed and formally reject the work order to prevent it from flowing into the next node and causing possible adverse consequences. At the same time as the rejection, a detailed risk report is usually attached so that the work order submitter can understand the problem and make corresponding modifications or supplements according to the guidance before resubmitting it for review.

[0114] S708. Perform quality inspection on the risk audit big model according to the preset period, and update the configuration of the risk audit big model according to the quality inspection results.

[0115] Optionally, in order to ensure that the risk audit model can continue to efficiently and accurately serve the risk management needs of the enterprise, a systematic quality inspection process can also be set for the risk audit model. This process is executed according to a pre-planned cycle, aiming to comprehensively evaluate and optimize the performance of the model. This periodic quality inspection is not only a monitoring of the current operating status of the model, but also an early warning and prevention of potential risks in the future. Among them, the preset period can be one day, one week, one month, etc., and this specification embodiment is not limited to this.

[0116] Furthermore, the quality inspection process covers two aspects: audit result quality inspection and model architecture quality inspection, but is not limited to this. During the specific implementation, it can be flexibly adjusted according to the actual situation to ensure that all key performance of the risk audit large model is covered.

[0117] On the one hand, for the quality inspection of audit results, the focus is mainly on the output quality of the big model in practical applications. By comparing the audit results output by the big model with manual audits or known standard answers, the accuracy, consistency and stability of the model are evaluated. The specific steps include: (1) Sample selection: Randomly or according to specific rules, a certain number of samples are selected from the recently processed pending audit work orders as the test set. (2) Result comparison: The model's audit results for these samples are compared one by one with the manual audit results or the correct answers verified in history. (3) Performance analysis: Based on the comparison results, the model's accuracy, recall rate, F1 score and other key indicators are calculated, and the performance differences of the model in identifying different types of risks are analyzed. (4) Anomaly detection: Identify and record anomalies or inconsistencies in the model output to provide a basis for subsequent problem location and improvement.

[0118] On the other hand, for the model architecture quality inspection, the focus is on the rationality, effectiveness and maintainability of the model's internal structure and parameters. Quality inspection methods include but are not limited to code review, parameter tuning, architecture evaluation, etc.

[0119] Optionally, the configuration of the risk audit model can be further updated based on the quality inspection results. This may include adjusting the model's parameter settings, optimizing the algorithm logic, updating the training data, and even reconstructing the model architecture to ensure that the model can continue to adapt to the changing transaction environment and risk characteristics, and provide enterprises with more accurate and efficient risk management services.

[0120] In the embodiments of this specification, a risk audit method for data work orders is provided, work orders that are judged to be at risk are re-inspected, a stable variety of audit methods are used in the re-inspection nodes, and the work orders are finally processed according to the re-inspection results. The risks are strictly managed throughout the process, and the efficiency and security of the transaction process are ensured through clear node division and result processing mechanisms, providing a solid guarantee for the safe operation of the system. In addition, the large model is quality inspected on a periodic basis, and the large model is updated and adjusted according to the quality inspection results to ensure that the model can continue to adapt to the changing application environment and risk characteristics, and provide enterprises with more accurate and efficient risk management services.

[0121] See also Figure 8 , Figure 8 This is a structural block diagram of a risk audit device for a data work order provided in an embodiment of this specification. Figure 8 As shown, the risk audit device 800 for data work orders includes:

[0122] The data acquisition module 810 is used to extract feature data from the work order to be reviewed;

[0123] Model review module 820, used to input feature data into the risk review big model, and determine the risk review result of the to-be-reviewed work order output by the risk review big model based on the feature data;

[0124] The review pass module 830 is used to confirm that the work order to be reviewed has passed the review if the risk review result is that there is no risk;

[0125] The risk marking module 840 is used to mark the work order to be reviewed as a risky work order if the risk review result is that there is a risk.

[0126] Optionally, the model review module 820 is also used to determine the risk review results of the work order to be reviewed after the risk review big model analyzes the feature data according to at least one prompt word; the prompt word is obtained after configuration based on at least one type of risk information, and the prompt word is used to specify the recognition purpose and output rules of the risk review big model.

[0127] Optionally, the types of risk information include at least one of content authenticity risk, content and product relevance risk, content security risk, and intellectual property risk; when there are at least two risk types, each prompt word is configured based on a different risk type.

[0128] Optionally, when there are at least two prompt words, the risk review model is configured to divide the at least two prompt words into multiple prompt word batches, and call each prompt word batch in sequence; wherein each prompt word batch includes at least one prompt word.

[0129] Optionally, the data acquisition module 810 is also used to extract text feature data of the text when it is recognized that the work order to be reviewed includes text; when it is recognized that the work order to be reviewed includes a picture, the picture is used as the picture feature data of the work order to be reviewed, and the picture text recognition is performed on the picture to obtain the text feature data of the picture; the picture text recognition method includes at least one of an optical character recognition method and a visual language model recognition method.

[0130] Optionally, the feature data is text feature data, and the data acquisition module 810 is also used to control the risk review model to call the text analysis model to analyze the text feature data and output the risk review result of the work order to be reviewed.

[0131] Optionally, the feature data is image feature data, and the data acquisition module 810 is also used to control the risk review model to call the visual language model to analyze the image feature data and output the risk review result of the work order to be reviewed.

[0132] Optionally, the risk review device 800 for data work orders also includes: a work order flow control module, which is used to determine whether the work order to be reviewed needs to be input into the risk review big model for review; if necessary, execute the step of inputting the characteristic data into the risk review big model; if not, transfer the work order to be reviewed to the manual review node.

[0133] Optionally, the work order flow control module is also used to determine whether the type of characteristic data in the work order to be reviewed meets the calling conditions of the risk review big model; if not, it is determined that the work order to be reviewed does not need to be input into the risk review big model for review.

[0134] Optionally, the work order flow control module is also used to perform pre-filtering processing on the work orders to be reviewed, and the pre-filtering processing includes at least one processing method of keyword matching and historical data deduplication; if the work order to be reviewed passes the pre-filtering processing, it is determined that the work order to be reviewed needs to be input into the risk review big model for review; if the work order to be reviewed does not pass the pre-filtering processing, it is determined that the work order to be reviewed does not need to be input into the risk review big model for review, and the work order to be reviewed is transferred to the re-inspection node.

[0135] Optionally, the risk review device 800 for data work orders also includes: a risk review module, which is used to transfer the risk work order to a review node, and determine the review result of the risk work order in the review node; if the review result is that there is no risk, the work order to be reviewed is confirmed to have passed the review; if the review result is that there is a risk, the work order to be reviewed will be intercepted and rejected.

[0136] Optionally, the review node includes at least one of a manual review node and a neural network model review node.

[0137] Optionally, the risk review device 800 for the data work order also includes: a model quality inspection module, which is used to perform quality inspection on the risk review large model according to a preset period, and update the configuration of the risk review large model according to the quality inspection results; the quality inspection includes at least one of the audit result quality inspection and the model architecture quality inspection.

[0138] Optionally, the work order to be reviewed is a marketing advertising work order.

[0139] See also Fig. 9 , Fig. 9 This is a schematic diagram of the structure of a terminal provided in an embodiment of this specification. Fig. 9 As shown, the terminal 900 may include: at least one terminal processor 901 , at least one network interface 904 , a user interface 903 , a memory 905 , and at least one communication bus 902 .

[0140] The communication bus 902 is used to realize the connection and communication between these components.

[0141] The user interface 903 may include a display screen (Display), a camera (Camera), and optionally the user interface 903 may also include a standard wired interface and a wireless interface. The network interface 904 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).

[0142] Among them, the terminal processor 901 may include one or more processing cores. The terminal processor 901 uses various interfaces and lines to connect various parts within the entire terminal 900, and executes various functions and processes data of the terminal 900 by running or executing instructions, programs, code sets or instruction sets stored in the memory 905, and calling data stored in the memory 905. Optionally, the terminal processor 901 can be implemented in at least one hardware form of digital signal processing (Digital Signal Processing, DSP), field programmable gate array (Field-Programmable Gate Array, FPGA), and programmable logic array (Programmable Logic Array, PLA). The terminal processor 901 can integrate one or a combination of a central processing unit (Central Processing Unit, CPU), a graphics processing unit (Graphics Processing Unit, GPU) and a modem. Among them, the CPU mainly processes the operating system, user interface and application programs; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the terminal processor 901, and it can be implemented separately through a chip.

[0143] Among them, the memory 905 may include a random access memory (Random Access Memory, RAM) and may also include a read-only memory (Read-Only Memory, ROM). Optionally, the memory 905 includes a non-transitory computer-readable storage medium. The memory 905 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 905 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store data involved in the above-mentioned various method embodiments, etc. The memory 905 may also be optionally at least one storage device located away from the aforementioned terminal processor 901. As Fig. 9As shown, the memory 905 as a computer storage medium may include an operating system, a network communication module, a user interface module, and a risk review program for data work orders.

[0144] exist Fig. 9 In the terminal 900 shown, the user interface 903 is mainly used to provide an input interface for the user and obtain the data input by the user; and the terminal processor 901 can be used to call the risk review program of the data work order stored in the memory 905, and specifically perform the following operations:

[0145] Extract feature data from the work orders to be reviewed;

[0146] Input the characteristic data into the risk review model, and determine the risk review results of the work orders to be reviewed output by the risk review model based on the characteristic data;

[0147] If the risk review result is that there is no risk, the work order to be reviewed is confirmed to have passed the review;

[0148] If the risk review result is that there is a risk, the work order to be reviewed will be marked as a risky work order.

[0149] In some embodiments, when the terminal processor 901 determines the risk review result of the work order to be reviewed output by the risk review big model based on the feature data, it specifically performs the following steps: determine the risk review result of the work order to be reviewed output by the risk review big model after analyzing the feature data according to at least one prompt word; the prompt word is obtained after configuration based on at least one type of risk information, and the prompt word is used to specify the recognition purpose and output rules of the risk review big model.

[0150] In some embodiments, the types of risk information include at least one of content authenticity risk, content and product relevance risk, content security risk, and intellectual property risk; when there are at least two risk types, each prompt word is configured based on a different risk type.

[0151] In some embodiments, when there are at least two prompt words, the risk review model is configured to divide the at least two prompt words into multiple prompt word batches, and call each prompt word batch in sequence; wherein each prompt word batch includes at least one prompt word.

[0152] In some embodiments, when the terminal processor 901 extracts feature data from a work order to be reviewed, it specifically performs the following steps: when it is recognized that the work order to be reviewed includes text, the text feature data of the text is extracted; when it is recognized that the work order to be reviewed includes a picture, the picture is used as the picture feature data of the work order to be reviewed, and the picture text recognition is performed on the picture to obtain the text feature data of the picture; the method of picture text recognition includes at least one of an optical character recognition method and a visual language model recognition method.

[0153] In some embodiments, the feature data is text feature data. When the terminal processor 901 determines the risk review result of the work order to be reviewed based on the feature data output by the risk review big model, it specifically performs the following steps: control the risk review big model to call the text analysis model to analyze the text feature data, and output the risk review result of the work order to be reviewed.

[0154] In some embodiments, the feature data is image feature data. When the terminal processor 901 determines the risk review result of the work order to be reviewed based on the feature data output by the risk review big model, it specifically performs the following steps: control the risk review big model to call the visual language model to analyze the image feature data, and output the risk review result of the work order to be reviewed.

[0155] In some embodiments, after extracting feature data from the work order to be reviewed, the terminal processor 901 also specifically performs the following steps: determining whether the work order to be reviewed needs to be input into the risk review model for review; if necessary, executing the step of inputting the feature data into the risk review model; if not, transferring the work order to be reviewed to the manual review node.

[0156] In some embodiments, when the terminal processor 901 determines whether the work order to be reviewed needs to be input into the risk review big model for review, it specifically performs the following steps: determines whether the type of characteristic data in the work order to be reviewed meets the calling conditions of the risk review big model; if not, it is determined that the work order to be reviewed does not need to be input into the risk review big model for review.

[0157] In some embodiments, when the terminal processor 901 determines whether the work order to be reviewed needs to be input into the risk review big model for review, it specifically performs the following steps: pre-filtering the work order to be reviewed, and the pre-filtering process includes at least one processing method of keyword matching and historical data deduplication; if the work order to be reviewed passes the pre-filtering process, it is determined that the work order to be reviewed needs to be input into the risk review big model for review; if the work order to be reviewed does not pass the pre-filtering process, it is determined that the work order to be reviewed does not need to be input into the risk review big model for review, and the work order to be reviewed is transferred to the re-inspection node.

[0158] In some embodiments, after marking the work order to be reviewed as a risky work order, the terminal processor 901 further specifically performs the following steps: transferring the risky work order to the re-inspection node, and determining the re-inspection result of the risky work order in the re-inspection node; if the re-inspection result is that there is no risk, confirming that the work order to be reviewed has passed the review; if the re-inspection result is that there is a risk, intercepting and rejecting the work order to be reviewed.

[0159] In some embodiments, the review node includes at least one of a manual review node and a neural network model review node.

[0160] In some embodiments, the terminal processor 901 also specifically performs the following steps: performing quality inspection on the risk audit big model according to a preset period, and updating the configuration of the risk audit big model according to the quality inspection results; the quality inspection includes at least one of the audit result quality inspection and the model architecture quality inspection.

[0161] In some embodiments, the work order to be reviewed is a marketing advertisement work order.

[0162] The embodiments of this specification provide a computer program product including instructions, and when the computer program product is run on a computer or a processor, the computer or the processor executes the steps of the method in any one of the above embodiments. The embodiments of this specification also provide a computer storage medium, which can store multiple instructions, and the instructions are suitable for being loaded by a processor and executing the steps of the method in any one of the above embodiments. Among them, the device, computer-readable storage medium, computer program product or chip provided in the embodiments of the present application are all used to execute the corresponding methods provided above, so the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding methods provided above, and will not be repeated here.

[0163] In the several embodiments provided in this specification, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of modules is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0164] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0165] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The above computer program product includes one or more computer instructions. When the above computer program instructions are loaded and executed on a computer, the above process or function according to the embodiment of this specification is generated in whole or in part. The above computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The above computer instructions can be stored in a computer-readable storage medium or transmitted by the above computer-readable storage medium. The above computer instructions can be transmitted from a website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (Digital Subscriber Line, DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode. The above computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server, data center, etc. that contains one or more available media integrated. The above-mentioned available media can be magnetic media (for example, floppy disks, hard disks, tapes), optical media (for example, digital versatile discs (DVD)), or semiconductor media (for example, solid state drives (SSD)), etc.

[0166] It should be noted that, for the convenience of description, the aforementioned method embodiments are all described as a series of action combinations, but those skilled in the art should be aware that this specification is not limited by the order of the actions described, because according to this specification, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by this specification.

[0167] In addition, it should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.) and signals involved in the embodiments of this specification are all authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions. For example, the marketing advertisements and work orders to be reviewed involved in the embodiments of this specification are all obtained with full authorization.

[0168] The above is a description of a specific embodiment of the present specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0169] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0170] The above is a description of a risk review method, device, storage medium and terminal for a data work order provided in an embodiment of this specification. For technical personnel in this field, according to the ideas of the embodiments of this specification, there may be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on this application.

Claims

1. A risk audit method for a data work order, characterized in that: The method comprises: Extract feature data from the work orders to be reviewed; Inputting the characteristic data into a risk review model, and determining a risk review result of the work order to be reviewed output by the risk review model based on the characteristic data; If the risk review result is that there is no risk, then the work order to be reviewed is confirmed to have passed the review; If the risk review result is that there is a risk, the work order to be reviewed will be marked as a risky work order.

2. The method according to claim 1, characterized in that The step of determining the risk review result of the work order to be reviewed output by the risk review big model based on the feature data includes: Determine the risk review result of the work order to be reviewed outputted by the risk review big model after analyzing the feature data according to at least one prompt word; The prompt words are obtained after configuration based on at least one type of risk information, and the prompt words are used to specify the identification purpose and output rules of the risk review model.

3. The method according to claim 2, characterized in that The types of risk information include at least one of content authenticity risk, content and product relevance risk, content security risk, and intellectual property risk; when there are at least two risk types, each prompt word is configured based on a different risk type.

4. The method according to claim 2 or 3, characterized in that: When there are at least two prompt words, the risk review model is configured to divide the at least two prompt words into a plurality of prompt word batches, and call each prompt word batch in sequence; wherein each prompt word batch includes at least one prompt word.

5. The method according to claim 1, characterized in that The feature data extracted from the work order to be reviewed includes: When it is identified that the work order to be reviewed includes text, extracting text feature data of the text; When it is identified that the work order to be reviewed includes a picture, the picture is used as the picture feature data of the work order to be reviewed, and picture text recognition is performed on the picture to obtain text feature data of the picture; the picture text recognition method includes at least one of an optical character recognition method and a visual language model recognition method.

6. The method according to claim 5, characterized in that The characteristic data is text characteristic data, and the step of determining the risk review result of the to-be-reviewed work order output by the risk review big model based on the characteristic data includes: The risk review model is controlled to call a text analysis model to analyze the text feature data, and output the risk review result of the work order to be reviewed.

7. The method according to claim 5, characterized in that The feature data is image feature data, and the step of determining the risk review result of the to-be-reviewed work order output by the risk review big model based on the feature data includes: The risk review model is controlled to call the visual language model to analyze the image feature data, and output the risk review result of the work order to be reviewed.

8. A computer program product comprising instructions, which, when executed on a computer or a processor, causes the computer or the processor to execute the steps of the method according to any one of claims 1 to 7.

9. A computer storage medium storing a plurality of instructions, wherein the instructions are suitable for being loaded by a processor and executing the steps of the method according to any one of claims 1 to 7.

10. A terminal comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method according to any one of claims 1 to 7 when executing the computer program.

Images