Audit method and system applied to advertisement image material

Abstract

The invention relates to the technical field of advertisement auditing, in particular to an auditing method and system applied to advertisement image materials. The method comprises the following steps: screening out a to-be-manually audited material through model analysis, and obtaining an audit level of the to-be-manually audited material according to a risk confidence coefficient; obtaining an intelligent auditing difficulty index according to the risk confidence of the reflected violation risk type of the reflected violation material in the to-be-manually audited material and all risk confidence of the non-reflected violation material; considering the condition of the reflected violation material corresponding to the uploading user and the risk degree of the reflected violation risk type, and obtaining the potential violation possibility; and in combination with the auditing level, the intelligent auditing difficulty index and the potential violation possibility, obtaining the auditing priority to perform manual auditing arrangement. According to the method, the advertisement materials to be audited are comprehensively subjected to audit priority level evaluation according to the difficulty condition of intelligent audit and the risk condition of uploading the user, so that better manual audit arrangement is provided to improve the audit efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of advertisement review, and in particular to a review method and system applied to advertisement image materials. Background Art

[0002] The review of advertising image materials is a crucial part of the advertising industry, mainly aimed at ensuring that advertising content complies with relevant laws, regulations and industry standards. With the rapid development of digital advertising, the review process has become more complicated, requiring not only review of image content, but also consideration of cultural sensitivity, intellectual property rights, and brand image. Currently, many companies are gradually introducing artificial intelligence technology to improve review efficiency and accuracy, but there are still some advertisers who use tricks to pass intelligent review, so manual review is still needed to ensure the compliance and appropriateness of the final content.

[0003] As consumers' expectations for advertising content increase and the number of people evading audit standards increases, it is increasingly difficult for existing artificial intelligence technology to achieve accurate audit results, and manual audit assistance is needed. However, manual audit time is limited, and different audit materials have different difficulty requirements. Improper arrangement of manual audit materials will also lead to low efficiency of advertising audit. Summary of the invention

[0004] In order to solve the technical problem in the prior art that the unreasonable arrangement of manual review materials leads to low efficiency of advertising review, the purpose of the present invention is to provide a review method and system for advertising image materials. The technical solution adopted is as follows: The present invention provides a review method for advertising image materials, the method comprising: Acquire the advertising materials to be reviewed in the scheduled time period and the corresponding uploading users; obtain the risk confidence of the advertising materials to be reviewed under each risk type through a pre-trained risk estimation model; the advertising materials to be reviewed include non-reported violation materials and reported violation materials, and the reported violation materials include reported violation risk types; Screen out materials to be manually reviewed based on the reported violations and risk confidence of the advertising materials to be reviewed; and obtain the review level of each material to be manually reviewed based on the risk confidence; According to the risk confidence of the reported illegal materials in the materials to be manually reviewed in the reported illegal risk types and the risk confidence of the non-reported illegal materials in all risk types, the intelligent review difficulty index of each material to be manually reviewed is obtained; Based on the distribution of the number of reported illegal materials by the users who uploaded the materials to be manually reviewed, and the high risk level of the uploaded users in the reported illegal risk types, the potential illegality of the materials to be manually reviewed is obtained; According to the review level, intelligent review difficulty index and potential violation possibility of each material to be reviewed, the review priority of each material to be reviewed is obtained; manual review is arranged according to the review priority of all materials to be reviewed in the scheduled time period.

[0005] Furthermore, the method for obtaining the material to be manually reviewed includes: All the materials reported to be in violation of regulations in the advertising materials to be reviewed shall be treated as materials to be manually reviewed; For any non-reported illegal material in any advertising material to be reviewed, the maximum risk confidence of the non-reported illegal material is used as a screening indicator; When the screening index is greater than the preset lower threshold and less than the preset upper threshold, the corresponding non-reported illegal materials will be treated as materials to be manually reviewed; the preset lower threshold is less than the preset upper threshold.

[0006] Furthermore, the method for obtaining the audit level includes: For any material to be manually reviewed, the maximum value of all risk confidence levels of the material to be manually reviewed is used as the evaluation value of the material to be manually reviewed; The difference between the evaluation value of the material to be manually reviewed and the preset lower limit threshold is negatively correlated and normalized to obtain the review level of the material to be manually reviewed.

[0007] Furthermore, the method for obtaining the intelligent audit difficulty index includes: For any material to be manually reviewed, the ratio between the maximum risk confidence of the material to be manually reviewed and the average of all risk confidences is used as the evaluation deviation of the material to be manually reviewed; When the material to be manually reviewed is not a material reported as violating regulations, the sum of all risk confidences of the material to be manually reviewed is used as the risk severity of the material to be manually reviewed; When the material to be manually reviewed is a material reported as violating regulations, the sum of the risk confidences of all the risk types of violations reported for the material to be manually reviewed is used as the risk severity of the material to be manually reviewed; Based on the assessment deviation and risk severity of the material to be manually reviewed, an intelligent review difficulty index for the material to be manually reviewed is obtained.

[0008] Furthermore, the possible methods of obtaining the potential violation include: For any material to be manually reviewed, the uploading user corresponding to the material to be manually reviewed is used as the analysis user; According to the proportion of reported illegal materials of the analyzed user and the quantity distribution of reported illegal risk types, the reported illegal degree index of the analyzed user is obtained; Obtain the reported violation index of the analyzed user according to the proportion of the risk confidence level under each reported violation risk type in the reported violation material of the analyzed user; By combining and analyzing the user's reported violation degree index and reported violation index, as well as the maximum risk confidence of the material to be manually reviewed, the potential violation possibility of the material to be manually reviewed is obtained.

[0009] Furthermore, the method for obtaining the reflected violation degree indicator includes: Calculate the ratio of the total number of reported illegal materials among the analyzed user to all the pending materials corresponding to the analyzed user, and obtain the reported illegal ratio of the analyzed user; Calculate and analyze the ratio of the total number of different types of reported violation risk types corresponding to all reported violation materials in the user to the total number of all risk types to obtain the type ratio for analysis; The reported violation degree index of the analyzed user is obtained by combining the reported violation proportion and type proportion of the analyzed user.

[0010] Furthermore, the method for obtaining the reported violation indicators includes: For any reported violation risk type in the reported violation materials of the analyzed user, the sum of the risk confidences of the reported violation risk type in all the materials to be manually reviewed is calculated as the risk indicator of the reported violation risk type; Calculate the sum of all risk confidences of the analyzed user under the reported violation risk type as the high risk value of the analyzed user under the reported violation risk type; The ratio of the high risk value of the reported violation risk type to the risk index is used as the probability of the user violating the reported violation risk type; By combining the violation likelihood of the analyzed user corresponding to all reported violation risk types, the reported violation index of the analyzed user is obtained.

[0011] Furthermore, the method for obtaining the audit priority includes: The product of the review level, intelligent review difficulty index and potential violation possibility of each material to be manually reviewed is normalized to obtain the review priority of each material to be manually reviewed.

[0012] Furthermore, the manual review arrangement is performed according to the review priority of all materials to be manually reviewed in the scheduled time period, including: Arrange all materials to be manually reviewed in the scheduled time period in descending order of review priority to obtain a material arrangement sequence; The material arrangement sequence is divided by using the maximum inter-class variance method to obtain a high-level arrangement sequence and a primary arrangement sequence; the mean value of the audit priority in the high-level arrangement sequence is greater than the mean value of the audit priority in the low-level arrangement sequence; Arrange the materials to be manually reviewed in the advanced arrangement sequence to senior reviewers in order; arrange the materials to be manually reviewed in the primary arrangement sequence to junior reviewers in order.

[0013] The present invention also provides an audit system applied to advertising image materials, including a memory and a processor, wherein the processor executes a computing program stored in the memory to implement the above-mentioned audit method applied to advertising image materials.

[0014] The present invention has the following beneficial effects: The present invention screens out materials to be manually reviewed through the risk confidence of different risk types of the advertising materials to be reviewed under model analysis, and obtains the review level of the materials to be manually reviewed according to the risk confidence, and reflects the degree of attention required for manual identification through artificial intelligence identification of risk situations. Further, considering that some of the reviewed materials are materials that are reported to be in violation and need to be reviewed again, the intelligent review difficulty index is obtained through the risk confidence of the materials to be manually reviewed in the risk type of the reflected violation and all the risk confidence of the materials not to be reflected in the violation, considering that the materials to be manually reviewed have also passed an audit, so the materials not to be reflected in the violation pass all identification situations, and the materials to be reflected in the violation pass the intelligent identification situation of the risk type of the reflected violation, reflecting the difficulty evaluation of the intelligent audit results for subsequent audits. Further, from the perspective of the upload user corresponding to the materials to be manually reviewed, the potential violation risks are analyzed, and the situation of the upload user corresponding to the materials to be manually reviewed and the risk level of the materials to be manually reviewed in the risk type of the reflected violation are considered, and the potential violation possibility is obtained, which characterizes the possibility of violation of the upload user corresponding to the materials to be manually reviewed. Finally, the audit priority is combined with the audit level, the intelligent audit difficulty index and the potential violation possibility, and the manual audit arrangement is arranged through the audit priority to optimize the arrangement strategy of the manual audit. The present invention comprehensively evaluates the review priority of the advertising materials to be reviewed by considering the difficulty of intelligent review and the risk of the uploading user, thereby providing a better manual review arrangement to improve the review efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 A flow chart of a review method for advertising image materials provided by an embodiment of the present invention; Figure 2 A flow chart of a method for obtaining potential violations provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, features and effects of a review method and system for advertising image materials proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0018] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0019] The following is a detailed description of a specific scheme of a review method and system for advertising image materials provided by the present invention in conjunction with the accompanying drawings.

[0020] See also Figure 1 , which shows a flow chart of a review method for advertising image materials provided by an embodiment of the present invention, the method comprising the following steps: S1: Obtain the advertising materials to be reviewed in the scheduled time period and the corresponding uploading users; obtain the risk confidence of the advertising materials to be reviewed under each risk type through a pre-trained risk estimation model; the advertising materials to be reviewed include non-reported violation materials and reported violation materials, and the reported violation materials include reported violation risk types.

[0021] For advertising image materials, there is a high possibility that there is a lot of obscure violation information. The sensitivity of artificial intelligence identification usually cannot meet the review requirements. Therefore, it is necessary to combine manual review assistance. In an embodiment of the present invention, the advertising materials to be reviewed in the scheduled time period and the uploading users who uploaded the advertising materials are obtained, and the risk type of violation involving the material is first evaluated through an intelligent model to obtain the risk confidence of different risk types.

[0022] In a specific embodiment of the present invention, through time limit, all advertising materials to be reviewed in each scheduled time period are reviewed and analyzed, and the scheduled time period can be set to 3 hours. In other embodiments of the present invention, restrictions can also be imposed based on quantity. When the number of advertising materials to be reviewed reaches a quantity threshold, all advertising materials to be reviewed are reviewed and analyzed. It should be noted that the time period and quantity threshold can be adjusted by the implementer according to the specific implementation scenario, and are not limited here.

[0023] Among them, the risk type refers to different violations that may cause adverse effects in the advertising material. The risk confidence level refers to the possibility of the risk type existing in the material. The greater the risk confidence level, the more likely it is that the corresponding risk type exists.

[0024] In an embodiment of the present invention, the input of the pre-trained risk assessment model is advertising image material, and the output is the risk confidence of each risk type. As an example, a convolutional neural network is selected as the basis of the model, and advertising image materials marked with different risk type labels are trained for recognition. During the training process, data enhancement technology is used to improve the generalization ability of the model. When estimating the risk confidence, the output probability of the softmax function can be used as the confidence. Use the validation set to evaluate the recognition ability of the model and monitor the performance indicators of the model, such as accuracy, recall rate, and F1 score. Adjust and optimize the model according to the evaluation results. It should be noted that the specific model training process is a technical means well known to those skilled in the art, and is not limited or elaborated here.

[0025] The advertising creatives to be reviewed include not only the advertising creatives that have been initially reviewed. Advertising creatives that have been reported to have a higher level of violations need to re-enter the review process. Therefore, the advertising creatives to be reviewed include non-reported violation materials and reported violation materials. At the same time, when a violation is reported, the advertising creative needs to select a risk type. Therefore, each reported violation material will have a corresponding reported violation risk type, and there may be multiple reported violation risk types.

[0026] S2: Screen out materials to be manually reviewed based on the reported violations and risk confidence of the advertising materials to be reviewed; and obtain the review level of each material to be manually reviewed based on the risk confidence.

[0027] For the advertising materials to be reviewed, they need to be screened by the intelligent model first. For the advertising materials to be reviewed that are relatively clear about the risk of violation or not, the review can be completed directly, but for the middle part with strong uncertainty, manual review assistance is required. In addition, for the advertising materials that are reported to have violated the regulations, it means that the previous review results are inappropriate and need to be further inspected manually. Therefore, the materials to be reviewed that need manual assistance are screened out based on the reported violations and risk confidence.

[0028] Preferably, in an embodiment of the present invention, the method for obtaining materials to be manually reviewed includes: First, all the materials reported to be in violation of regulations among the pending advertising materials will be treated as materials to be manually reviewed. For the materials reported to be in violation of regulations that need to be re-reviewed, it means that the initial review does not meet the requirements and further manual inspection is required.

[0029] Secondly, for any non-reported violation material in the pending advertising material, the maximum risk confidence of the non-reported violation material is used as a screening indicator. The maximum value of the risk confidence can reflect the most likely degree of the risk type of the advertising material. When the screening indicator is large, it means that the advertising material is a material with a significant corresponding risk type, and it can be directly intelligently reviewed as not passed. When the screening indicator is small, it means that the possibility of the most likely risk type in the advertising material is also extremely low, and it can be directly intelligently reviewed as passed, and the need for manual inspection is low.

[0030] Therefore, when the screening index is greater than the preset lower threshold and less than the preset upper threshold, it indicates that the most likely risk type of the corresponding advertising material is not significantly present, nor is there a very low possibility, and the corresponding non-reported illegal material is treated as material to be manually reviewed. In the embodiment of the present invention, the preset lower threshold is less than the preset upper threshold, so that the judgment can be realized. The preset lower threshold can be set to 0.2, and the preset upper threshold can be set to 0.8. The specific values ​​can be adjusted by the implementer, and are not limited here.

[0031] Further preliminary assessment is conducted on each material to be manually reviewed to reflect the degree of need for manual review. The greater the risk confidence of the most likely risk type of the material to be manually reviewed, the more obvious the existence of the risk type is. Conversely, the more vague the judgment on the existence of the most likely risk type is, the higher the demand for further manual analysis is.

[0032] Preferably, in an embodiment of the present invention, the method for obtaining the audit level includes: For any material to be manually reviewed, the maximum value of all risk confidence levels of the material to be manually reviewed is used as the evaluation value of the material to be manually reviewed, and the significance of the risk type is analyzed based on the maximum risk confidence level at this time.

[0033] The difference between the evaluation value of the material to be manually reviewed and the preset lower limit threshold is further negatively correlated and normalized to obtain the review level of the material to be manually reviewed. The closer the evaluation value is to the preset lower limit threshold, the more vague the judgment of the most likely risk type is, so the higher the review level needs to be, the higher the need for manual review.

[0034] It should be noted that negative correlation mapping and normalization processing are both technical means well known to technical personnel in this field. Negative correlation mapping can be in the form of inverse proportional value or negative exponential power, and normalization can be linear normalization or standard normalization, etc. The specific method is not limited here.

[0035] At this point, the screening and needs assessment of materials that require manual review are completed.

[0036] S3: Obtain the intelligent review difficulty index of each material to be manually reviewed based on the risk confidence of the reported violation risk type of the material to be manually reviewed and the risk confidence of the non-reported violation risk type of the material to be manually reviewed.

[0037] For advertising materials that require manual review, the multiple risk type results of intelligent review reflect the degree of diversity of possible risk types in the materials. The more diverse the materials, the more overall assessment can be made during intelligent review analysis. Conversely, when the degree of diversity is weaker and the degree of single significance is not high, the subsequent judgment reflected by intelligent review is more difficult, the result accuracy is lower, and the requirements for manual review are higher. In addition, for materials reported as violating regulations, the results of the reported violation risk types can better reflect the judgment of intelligent review.

[0038] Preferably, in an embodiment of the present invention, the method for obtaining the intelligent audit difficulty index includes: First, for any material to be manually reviewed, the ratio between the maximum risk confidence of the material to be manually reviewed and the mean of all risk confidences is used as the evaluation deviation of the material to be manually reviewed, reflecting the single significance of the most likely risk type. The larger the evaluation deviation, the more significant the most likely risk type, and the simpler the judgment and analysis. Conversely, it means that the advertising material has multiple more significant risk types, and there is a high possibility of multiple violations, which may make the subsequent review and judgment easier.

[0039] Furthermore, when the material to be manually reviewed is not a material that has been reported as a violation, the sum of all risk confidences of the material to be manually reviewed is used as the risk severity of the material to be manually reviewed. The higher the overall risk severity, the easier it is to make subsequent judgments and analyses. When the material to be manually reviewed is a material that has been reported as a violation, the sum of all risk confidences of the material to be manually reviewed that has been reported as a violation risk type is used as the risk severity of the material to be manually reviewed. The intelligent review of the material that has been reported as a violation based on the type of violation risk that has been reported is analyzed. The greater the risk severity, the easier it is for the intelligent review to reflect subsequent judgments.

[0040] Finally, the intelligent review difficulty index of the material to be manually reviewed is obtained by combining the assessment deviation and risk severity of the material to be manually reviewed. In an embodiment of the present invention, the ratio of the assessment deviation to the risk severity of the material to be manually reviewed is used as the intelligent review difficulty index of the material to be manually reviewed. When the assessment deviation is larger and the risk severity is smaller, it means that the intelligent review result reflects a more single risk type and the less significant the characterization effect, the more difficult the subsequent judgment and analysis is, and the higher the demand for manual review is.

[0041] S4: According to the distribution of the number of materials reported as illegal by the users who uploaded the materials to be manually reviewed, and the high risk level of the users who uploaded the materials to be manually reviewed in the reported illegal risk types, the potential illegality of the materials to be manually reviewed is obtained.

[0042] From the perspective of the user who uploaded the ad material, when there is a high risk impact of the uploader in the materials to be manually reviewed, it means that this uploader has uploaded many illegal materials. At this time, the materials to be manually reviewed are likely to have artificially circumvented illegal information, so more attention should be paid to the subsequent manual review. Therefore, the potential violations of the manually reviewed materials are analyzed based on the situation that the uploader has been reported to have illegal materials, and the risk level of the review under the reported illegal risk type.

[0043] Preferably, in the embodiment of the present invention, the method for obtaining the potential violation may refer to Figure 2 , which shows a flow chart of a method for obtaining potential violations provided by an embodiment of the present invention, the method comprising the following steps: S401: For any material to be manually reviewed, the uploading user corresponding to the material to be manually reviewed is used as the analysis user; and the reported violation degree index of the analysis user is obtained according to the proportion of the reported violation materials of the analysis user and the quantity distribution of the reported violation risk types.

[0044] For each material to be manually reviewed, analysis is performed from the perspective of the corresponding uploading user. The corresponding uploading user is used as the analysis user. By analyzing the proportion of the user's reported illegal materials and the reported illegal types, the degree of frequent violations reported by the user is reflected.

[0045] Preferably, in an embodiment of the present invention, the method for obtaining the reflected violation degree indicator includes: First, calculate the ratio of the total number of reported illegal materials among the analyzed user to all the materials to be reviewed corresponding to the analyzed user, and obtain the reported illegal ratio of the analyzed user, which reflects the proportion of reported illegal materials uploaded by the analyzed user.

[0046] Secondly, calculate and analyze the ratio of the total number of types of different reported violation risk types corresponding to all reported violation materials among the users to the total number of all risk types, and obtain the type ratio for analysis, which reflects the prevalence of violations in the advertising materials uploaded by the analyzed users.

[0047] Finally, by combining the reported violation ratio and type ratio of the analyzed user, we can obtain the reported violation degree index of the analyzed user. The larger the reported violation ratio and type ratio are, the more materials with violations uploaded by the analyzed user are, and the wider the aspects involved in the violations are, the higher the potential badness of the analyzed user is.

[0048] S402: Obtain the reported violation index of the analyzed user according to the proportion of the risk confidence level under each reported violation risk type in the reported violation material of the analyzed user.

[0049] Further risk confidence analysis of the types of violation risks reported to the analyzed users shows that the higher the overall risk confidence ratio of the analyzed users under the types of violation risks reported, the more deeply the analyzed users are involved in this violation, and it is very likely that they are still testing the review status of this violation, so a higher level of manual review and attention is required.

[0050] Preferably, in an embodiment of the present invention, the method for obtaining the reflected violation indicator includes: First, for any reported violation risk type in the reported violation materials of the analyzed user, calculate the sum of the risk confidence levels of the reported violation risk type in all materials to be manually reviewed as the risk indicator of the reported violation risk type, reflecting the possible degree of violation of the materials to be manually reviewed in the reported violation risk type.

[0051] The sum of all risk confidences of the analyzed user under the reflected violation risk type is further calculated as the high risk value of the analyzed user under the reflected violation risk type, reflecting the possible violation degree of the analyzed user on the reflected violation risk type.

[0052] Further, the ratio of the high risk value of the reported violation risk type to the risk index is used as the possibility of violation of the analyzed user in the reported violation risk type. By analyzing the proportion of the possible violation degree of the user in the overall advertising materials, the tentative violation degree of the analyzed user in the reported violation risk type is reflected. The higher the possibility of violation, the more serious the violation of the analyzed user.

[0053] Finally, the violation probability of the analyzed user corresponding to all reported violation risk types is combined to obtain the reported violation index of the analyzed user. In an embodiment of the present invention, the sum of the violation probability of the analyzed user corresponding to all reported violation risk types is calculated and normalized to obtain the reported violation index of the analyzed user, which comprehensively reflects the severity of the analyzed user's involvement in different violation situations.

[0054] S403: The potential violation possibility of the material to be manually reviewed is obtained by analyzing the reported violation degree index and the reported violation violation index of the user and the maximum risk confidence of the material to be manually reviewed.

[0055] By analyzing the degree of violation reported and the severity of violation involved from the user's perspective of the material to be manually reviewed, and taking into account the significant risk type of the material to be manually reviewed itself, the possibility of potential violation in the material to be manually reviewed is reflected. In an embodiment of the present invention, the product of the user's reported violation degree index and the reported violation index, and the maximum risk confidence of the material to be manually reviewed is normalized to obtain the potential violation possibility of the material to be manually reviewed.

[0056] When the reported violation level index and reported violation violation index of the analyzed user are larger, it means that the risk level of the corresponding uploading user is higher and the attention required is higher. When the maximum risk confidence is larger, it means that the most likely risk type is more significant, and the possibility of the analyzed user trying to avoid violations is higher. Therefore, the manual review needs more attention, and the potential possibility of violations is greater.

[0057] S5: Obtain the review priority of each material to be reviewed based on its review level, intelligent review difficulty index, and potential violation possibility; arrange manual review based on the review priority of all materials to be reviewed in the scheduled time period.

[0058] A comprehensive analysis of the three aspects of each material to be manually reviewed is performed, combined with the review level, intelligent review difficulty index and potential violation possibility to obtain the review priority. In an embodiment of the present invention, the product of the review level, intelligent review difficulty index and potential violation possibility of each material to be manually reviewed is normalized to obtain the review priority of each material to be manually reviewed.

[0059] The greater the audit level, intelligent audit difficulty index and potential violation possibility, the higher the degree of manual review attention required for the material to be manually reviewed, and the higher the review priority. When the subsequent audit arrangements are made, it can be assigned to more senior auditors and given priority to improve the audit efficiency.

[0060] Therefore, in some possible implementations of the embodiments of the present invention, manual review arrangements are subsequently made according to the review priorities of all materials to be manually reviewed in the scheduled time period, including: First, all materials to be manually reviewed in the scheduled time period are arranged in descending order of review priority to obtain a material arrangement sequence, which reflects the priority arrangement sequence of the advertising materials.

[0061] In order to improve the efficiency of material review, the material arrangement sequence is divided by the maximum inter-class variance method to obtain a high-level arrangement sequence and a primary arrangement sequence. Through sequence division, advertising materials with different attention requirements can be arranged to more suitable reviewers. The mean review priority in the high-level arrangement sequence is greater than the mean review priority in the low-level arrangement sequence, and the materials to be manually reviewed in the high-level arrangement sequence require a higher degree of attention. It should be noted that the maximum inter-class variance method is a technical means well known to those skilled in the art and will not be elaborated here.

[0062] Finally, the materials to be manually reviewed in the advanced arrangement sequence are arranged in sequence to senior reviewers, and the materials to be manually reviewed in the primary arrangement sequence are arranged in sequence to junior reviewers, so as to realize the arrangement decision of the materials for manual review. The reviewers give the result of whether the review is passed or not through manual auxiliary judgment, so as to improve the review efficiency of the overall advertising materials in each time period.

[0063] In summary, the present invention screens out materials to be manually reviewed through the risk confidence of different risk types of the advertising materials to be reviewed under model analysis, and obtains the review level of the materials to be manually reviewed according to the risk confidence, and reflects the degree of attention required for manual identification through artificial intelligence identification of risk situations. Furthermore, considering that some of the reviewed materials are materials that are reported to be in violation and need to be reviewed again, the intelligent review difficulty index is obtained through the risk confidence of the materials reflected in the violation in the materials to be manually reviewed in the risk type of the violation and all the risk confidence of the non-reflected violation materials. Considering that the materials reflected in the violation have also passed an audit, the materials that are not reflected in the violation pass all identification situations, and the materials reflected in the violation pass the intelligent identification situation of the risk type of the violation, reflecting the difficulty assessment of the subsequent audit by the intelligent audit results. Furthermore, from the perspective of the uploading user corresponding to the materials to be manually reviewed, the potential violation risks are analyzed, and the uploading user is considered to correspond to the situation of the materials reflected in the violation and the risk level of the reflected violation risk type, and the potential violation possibility is obtained, which characterizes the violation possibility of the uploading user corresponding to the materials to be manually reviewed. Finally, the audit priority is obtained by combining the audit level, intelligent audit difficulty index and potential violation possibility, and manual audit arrangement is made according to the audit priority to optimize the manual audit arrangement strategy. The present invention comprehensively evaluates the review priority of the advertising materials to be reviewed by considering the difficulty of intelligent audit and the risk of the uploading user, and then provides a better manual audit arrangement to improve the audit efficiency.

[0064] The present invention also provides an audit system applied to advertising image materials, including a memory and a processor, wherein the processor executes a computing program stored in the memory to implement the above-mentioned audit method applied to advertising image materials.

[0065] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying 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.

[0066] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A review method for advertising image materials, characterized in that: The method comprises: Acquire the advertising materials to be reviewed in the scheduled time period and the corresponding uploading users; obtain the risk confidence of the advertising materials to be reviewed under each risk type through a pre-trained risk estimation model; the advertising materials to be reviewed include non-reported violation materials and reported violation materials, and the reported violation materials include reported violation risk types; Screen out materials to be manually reviewed based on the reported violations and risk confidence of the advertising materials to be reviewed; and obtain the review level of each material to be manually reviewed based on the risk confidence; According to the risk confidence of the reported illegal materials in the materials to be manually reviewed in the reported illegal risk types and the risk confidence of the non-reported illegal materials in all risk types, the intelligent review difficulty index of each material to be manually reviewed is obtained; Based on the distribution of the number of reported illegal materials by the users who uploaded the materials to be manually reviewed, and the high risk level of the uploaded users in the reported illegal risk types, the potential illegality of the materials to be manually reviewed is obtained; According to the review level, intelligent review difficulty index and potential violation possibility of each material to be reviewed, the review priority of each material to be reviewed is obtained; manual review is arranged according to the review priority of all materials to be reviewed in the scheduled time period.

2. A review method for advertising image materials according to claim 1, characterized in that: The method for obtaining the material to be manually reviewed includes: All the materials reported to be in violation of regulations in the advertising materials to be reviewed shall be treated as materials to be manually reviewed; For any non-reported illegal material in any advertising material to be reviewed, the maximum risk confidence of the non-reported illegal material is used as a screening indicator; When the screening index is greater than the preset lower threshold and less than the preset upper threshold, the corresponding non-reported illegal materials will be treated as materials to be manually reviewed; the preset lower threshold is less than the preset upper threshold.

3. A review method for advertising image materials according to claim 2, characterized in that: The method for obtaining the audit level includes: For any material to be manually reviewed, the maximum value of all risk confidence levels of the material to be manually reviewed is used as the evaluation value of the material to be manually reviewed; The difference between the evaluation value of the material to be manually reviewed and the preset lower limit threshold is negatively correlated and normalized to obtain the review level of the material to be manually reviewed.

4. The audit method for advertising image materials according to claim 1, characterized in that: The method for obtaining the intelligent audit difficulty index includes: For any material to be manually reviewed, the ratio between the maximum risk confidence of the material to be manually reviewed and the average of all risk confidences is used as the evaluation deviation of the material to be manually reviewed; When the material to be manually reviewed is not a material reported as violating regulations, the sum of all risk confidences of the material to be manually reviewed is used as the risk severity of the material to be manually reviewed; When the material to be manually reviewed is a material reported as violating regulations, the sum of the risk confidences of all the risk types of violations reported for the material to be manually reviewed is used as the risk severity of the material to be manually reviewed; Based on the assessment deviation and risk severity of the material to be manually reviewed, an intelligent review difficulty index for the material to be manually reviewed is obtained.

5. The audit method for advertising image materials according to claim 1, characterized in that: Possible methods of obtaining the potential violations include: For any material to be manually reviewed, the uploading user corresponding to the material to be manually reviewed is used as the analysis user; According to the proportion of reported illegal materials of the analyzed user and the quantity distribution of reported illegal risk types, the reported illegal degree index of the analyzed user is obtained; Obtain the reported violation index of the analyzed user according to the proportion of the risk confidence level under each reported violation risk type in the reported violation material of the analyzed user; By combining and analyzing the user's reported violation degree index and reported violation index, as well as the maximum risk confidence of the material to be manually reviewed, the potential violation possibility of the material to be manually reviewed is obtained.

6. A review method for advertising image materials according to claim 5, characterized in that: The method for obtaining the reflected violation degree indicator includes: Calculate the ratio of the total number of reported illegal materials among the analyzed user to all the pending materials corresponding to the analyzed user, and obtain the reported illegal ratio of the analyzed user; Calculate and analyze the ratio of the total number of different types of reported violation risk types corresponding to all reported violation materials in the user to the total number of all risk types to obtain the type ratio for analysis; The reported violation degree index of the analyzed user is obtained by combining the reported violation proportion and type proportion of the analyzed user.

7. A review method for advertising image materials according to claim 5, characterized in that: The method for obtaining the reported violation indicators includes: For any reported violation risk type in the reported violation materials of the analyzed user, the sum of the risk confidences of the reported violation risk type in all the materials to be manually reviewed is calculated as the risk indicator of the reported violation risk type; Calculate the sum of all risk confidences of the analyzed user under the reported violation risk type as the high risk value of the analyzed user under the reported violation risk type; The ratio of the high risk value of the reported violation risk type to the risk index is used as the probability of the user violating the reported violation risk type; The violation probability of the analyzed user corresponding to all reported violation risk types is combined to obtain the reported violation index of the analyzed user.

8. The audit method for advertising image materials according to claim 1, characterized in that: The method for obtaining the audit priority includes: The product of the review level, intelligent review difficulty index and potential violation possibility of each material to be manually reviewed is normalized to obtain the review priority of each material to be manually reviewed.

9. The audit method for advertising image materials according to claim 1, characterized in that: The manual review arrangement is performed according to the review priority of all materials to be manually reviewed in the scheduled time period, including: Arrange all materials to be manually reviewed in the scheduled time period in descending order of review priority to obtain a material arrangement sequence; The material arrangement sequence is divided by using the maximum inter-class variance method to obtain a high-level arrangement sequence and a primary arrangement sequence; the mean value of the audit priority in the high-level arrangement sequence is greater than the mean value of the audit priority in the low-level arrangement sequence; Arrange the materials to be manually reviewed in the advanced arrangement sequence to senior reviewers in order; arrange the materials to be manually reviewed in the primary arrangement sequence to junior reviewers in order.

10. A review system for advertising image materials, comprising a memory and a processor, characterized in that: The processor executes the computing program stored in the memory to implement the review method for advertising image materials as described in any one of claims 1 to 9.

Images