Information inspection method, computer equipment and storage medium
By obtaining and analyzing multi-dimensional evaluation data in the information delivery system in real time, and combining the evaluation results, it automatically identifies the risk type of media information and triggers a secondary audit, which solves the problem of insufficient coverage and audit efficiency of information delivery systems in the existing technology, and realizes the Efficient and accurate screening and identification of violation content.
Patent Information
- Application Number
- CN201910420316.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2019-05-20
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2039-05-20
AI Technical Summary
When the existing information delivery system processes a large amount of media information, it is difficult to balance the coverage and audit manpower, resulting in insufficient audit effect and efficiency, and the inability to effectively identify and filter out illegal content.
By obtaining multi-dimensional evaluation data of media information in real time from the information delivery system, determining the risk type of media information based on the evaluation results of each dimension, triggering automatic recommendation to the secondary review process to ensure accurate identification and screening of illegal content.
It improves the efficiency and quality of information inspection, ensures the coverage and accuracy of risk type assessment of media information, reduces the workload of manual review, and improves the review efficiency and effectiveness.
Smart Images

Figure CN111966915B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of computer technology, and in particular to an information inspection method, a computer device and a storage medium. Background Art
[0002] With the increasing development of information technology, in order to let more people know and buy the products or services they produce, media information is usually produced to promote the products or services, and the media information is displayed on the display positions of information delivery systems that are well known to the public, such as web pages and social networks. Through the information delivery system, media information can be quickly pushed to a large number of users, thereby achieving the purpose of rapid information promotion.
[0003] However, with the emergence of various media information on the Internet, some media information that is illegally disseminated may appear. The dissemination of these illegal media information will cause very bad effects, and will also have a bad impact on the reputation of the information delivery system that delivers these media information. Therefore, it is necessary to conduct patrol inspections and audits on the media information delivered to the information delivery system. The currently known patrol inspection method is to randomly select a certain proportion of media information and inject it into the audit system for manual review. With the rapid increase in the amount of media information delivered on the Internet, it is difficult to achieve a good balance between coverage and audit manpower through random inspections, resulting in insufficient audit results and efficiency. Summary of the invention
[0004] The embodiments of the present invention provide an information inspection method, a computer device and a storage medium, which can effectively improve the audit efficiency and quality.
[0005] The technical solution of the embodiment of the present invention is achieved as follows:
[0006] An information inspection method includes: acquiring evaluation data of at least one dimension of delivered media information from an information delivery system in real time, determining an evaluation result of the media information corresponding to the at least one dimension based on the evaluation data; determining a risk type of the media information in combination with the evaluation results corresponding to each of the dimensions; and triggering the information delivery system to stop delivering the media information when the media information is of a set risk type.
[0007] An information inspection device includes: an evaluation module, which is used to obtain evaluation data of at least one dimension of the delivered media information from an information delivery system in real time, and determine the evaluation result of the media information corresponding to the at least one dimension based on the evaluation data; a risk determination module, which is used to determine the risk type of the media information in combination with the evaluation results corresponding to each of the dimensions; and a recommendation review module, which is used to trigger the information delivery system to stop delivering the media information when the media information is of a set risk type.
[0008] A computer device comprises a processor and a memory for storing a computer program that can be run on the processor; when the processor is used to run the computer program, the information inspection method described in the embodiment of the present invention is implemented.
[0009] A storage medium stores a computer program, which, when executed by a processor, implements the information inspection method described in an embodiment of the present invention.
[0010] In an embodiment of the present invention, the evaluation data of each dimension of the delivered media information is obtained in real time from the information delivery system to determine the evaluation result of the corresponding dimension, and the risk type of the media information is determined in combination with the evaluation results of each dimension. Therefore, the media information that may contain illegal content can be efficiently and accurately screened out according to the set dimensions, triggering the secondary review process of the media information that may contain illegal content. In this way, the media information of the set risk type can be comprehensively and objectively determined according to each dimension, ensuring the coverage of the assessment of the risk type of the media information, so as to improve the review efficiency and review quality. BRIEF DESCRIPTION OF THE DRAWINGS
[0011] Figure 1 A schematic diagram of an application scenario for implementing the information inspection method provided in an embodiment of the present invention;
[0012] Figure 2 A system framework diagram of an information inspection method in an embodiment of the present invention;
[0013] Figure 3 Flow chart of the information inspection method in an embodiment of the present invention;
[0014] Figure 4 Flow chart of the information inspection method in an embodiment of the present invention;
[0015] Figure 5 Schematic diagram of dimensions of media information and corresponding weight parameters in the information inspection method in an embodiment of the present invention;
[0016] Figure 6 Schematic diagram of the architecture of a combined model of a linear model and a neural network in an embodiment of the present invention;
[0017] Figure 7 A schematic diagram of an interface for providing feedback based on a user feedback entry in an embodiment of the present invention;
[0018] Figure 8 is a schematic diagram of displaying the media information according to the evaluation result in an embodiment of the present invention;
[0019] Fig. 9 Flow chart of the information inspection method in an embodiment of the present invention;
[0020] Fig.10 This is a schematic diagram of the structure of an information inspection device provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0021] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.
[0023] The embodiment of the present invention provides an information inspection method, an information inspection device for implementing the information inspection method, a computer device, and a storage medium storing an executable computer program for implementing the information inspection method. Among them, the implementation side of the information inspection method in the embodiment of the present invention can be a terminal and / or a server. The exemplary implementation scenario of the information inspection method in the embodiment of the present invention is described below.
[0024] like Figure 1 As shown, a schematic diagram of an optional application scenario for implementing the information inspection method provided in the embodiment of the present invention is provided. The architecture of the application scenario includes a first server 100, a second server 200 and a user terminal 300. In the embodiment of the present invention, the description of the application scenario of the information inspection method is described by taking the media information as an advertisement as an example. It should be noted that the advertisements in the following description can be replaced by media information. Please refer to Figure 2 , is a system framework diagram of the information inspection method provided by an embodiment of the present invention, the system framework includes an online advertising system, a recommendation-based inspection system and an audit system. The implementation side of the online advertising system, the recommendation-based inspection system and the audit system can respectively include one or more independent servers or terminals. As an optional embodiment, Figure 1As shown, the implementation side of the online advertising system can be the first server 100, the implementation side of the inspection system based on recommendation can be the second server 200, and the implementation side of the audit system can be the user terminal 300. The online advertising system is used to deliver and display the media information in the information delivery system. The inspection system based on recommendation is used to obtain the evaluation data of each set dimension of the media information currently being delivered from the information delivery system, determine the evaluation results of the media information in the corresponding dimension according to the evaluation data, and determine that the media information is a high-risk type of media information that may contain illegal content in combination with the evaluation results corresponding to each dimension and the corresponding weight parameters, triggering the online advertising system to suspend the media information offline and recommend the media information to the audit system. The audit system is used to receive the media information recommended by the inspection system based on recommendation, and conduct a secondary audit manually. When the secondary audit confirms that the media information belongs to the high-risk type of media information, the online advertising system is triggered to permanently offline the media information.
[0025] See also Figure 3, which is a flow chart of an information inspection method provided in an embodiment of the present invention, the information inspection method comprises the following steps: S11, the online advertising system displays the advertisements placed by the advertiser in the information delivery system; wherein the information delivery system may refer to a platform that is familiar to the public and can display media information, such as a web page and a social network. S12, based on the recommended inspection system, obtains the evaluation data of the set dimensions of the advertisements currently being displayed in the information delivery system, and determines the evaluation results of the advertisements in the corresponding dimensions according to the evaluation data; wherein, different dimensions can be set according to the degree of necessity of reviewing whether the advertisements contain illegal content, such as according to the communication influence of the corresponding ad slot of the advertisement, the degree of necessity of reviewing whether the advertisements on the ad slot with greater communication influence contain illegal content is correspondingly higher; according to the communication influence of the carrier to which the advertisement belongs, the degree of necessity of reviewing whether the advertisements on the ad carrier with greater communication influence contain illegal content is correspondingly higher; according to the conversion rate of the advertisement during the delivery cycle, the greater the conversion rate increment, the higher the degree of necessity of reviewing whether the advertisements contain illegal content; according to the type of users of the advertisement delivery, the more black seed type users among the users who receive the advertisement, the higher the degree of necessity of reviewing whether the advertisement contains illegal content; according to the industry involved in the advertisement content, the industry that needs to be reviewed is the industry that needs to be reviewed, and the degree of necessity of reviewing whether the advertisement contains illegal content is correspondingly higher. S13, the inspection system based on recommendations determines the high-risk type of advertisements that may contain illegal content according to the evaluation results corresponding to each of the dimensions. S14, recommend the high-risk type of advertisements to the review system; wherein, the inspection system based on recommendations sends the high-risk type of advertisement recommendations to the review system, and may also include triggering the online advertising system to temporarily take the advertisement offline. S15, the review system receives the high-risk type of advertisements recommended by the inspection system based on recommendations, and performs a secondary review of the advertisements received by the review system by manual review; wherein, when the secondary review confirms that the advertisement is a high-risk type of advertisement, it can trigger the online advertising system to permanently take the advertisement offline, and when the secondary review confirms that the advertisement is not a high-risk type of advertisement, it can trigger the online advertising system to restore the advertisement online to continue to complete the advertisement delivery plan.
[0026] See also Figure 4 , which is a flow chart of an information inspection method provided by an embodiment of the present invention, can be applied to Figure 1 The second server shown, it should be noted that the second server may also be other computer devices such as a terminal, which will be explained in conjunction with the following steps.
[0027] Step 101: obtaining evaluation data of at least one dimension of delivered media information from an information delivery system in real time.
[0028] An information delivery system refers to any one or more applications that can be used to display media information. Media information refers to text, images, and / or video information produced to promote a product or service in order to make more people aware of and purchase the product or service. Delivery refers to the process by which the provider of a product or service produces information recommending a product or server and displays it on a display position of an information delivery system that is well known to the public, such as a web page or social network. Among them, an information delivery system can refer to an application specifically used for delivering media information, such as an online shopping mall, etc.; it can also be an application that is not specialized but contains a designated functional module for delivering media information, such as a social application, video playback application, news application, etc. that includes a functional module for delivering media information.
[0029] Dimension refers to an evaluation angle that characterizes the risk type of media information based on different attributes of the media information. Among them, one dimension can correspond to an evaluation strategy, and each evaluation strategy is used to independently evaluate the degree of necessity of reviewing the illegal content contained in the media information based on the evaluation data of the corresponding dimension. The evaluation data of at least one dimension of the delivered media information obtained in real time from the information delivery system may include at least one of the following: obtaining evaluation data of at least one dimension of the media information in the current delivery plan in real time from the information delivery system; periodically obtaining evaluation data of at least one dimension of the media information currently being delivered and displayed from the information delivery system; obtaining evaluation data of at least one dimension of the media information in the schedule corresponding to the current time in real time according to the schedule of the media information delivered by the information delivery system.
[0030] Step 102: Determine an evaluation result of the media information corresponding to the at least one dimension according to the evaluation data.
[0031] Among them, dimension refers to a predetermined evaluation angle used to evaluate whether it is necessary to review the illegal content contained in the corresponding media information. One dimension corresponds to one evaluation result, which refers to the evaluation result of independently evaluating the degree of necessity of reviewing the illegal content contained in the media information based on the evaluation data of the corresponding dimension. Taking the media information as advertisements as an example, the dimensions of whether to review the illegal content contained in the corresponding advertisements may include: whether the ad slot where the advertisement is placed is an ad slot with a wide distribution range, the conversion evaluation dimension of whether the real-time conversion data of the advertisement is abnormal, the user evaluation dimension of whether the user or user group receiving the advertisement is abnormal, the advertising carrier evaluation dimension of whether the traffic of the carrier corresponding to the advertisement is abnormal, and the industry evaluation dimension of whether the industry to which the advertisement belongs complies with the advertising delivery rules.
[0032] The evaluation data of the dimension refers to the relevant data that determines the risk status of the corresponding media information in the corresponding dimension. Taking the media information as advertisement as an example, the evaluation data corresponding to the ad slot evaluation dimension includes the exposure data of the advertisement in the ad slot, the evaluation data corresponding to the conversion evaluation dimension includes the click volume data of the advertisement in two adjacent time periods, the evaluation data corresponding to the ad carrier evaluation dimension includes the visit volume data of the corresponding ad carrier in two adjacent time periods, the evaluation data corresponding to the user evaluation dimension includes the object attribute data that determines whether the user who clicks the advertisement belongs to the black seed user group, and the evaluation data corresponding to the industry evaluation dimension includes the industry attribute data that determines whether the advertisement content belongs to the specified industry.
[0033] Determining the evaluation result of the media information corresponding to the dimension according to the evaluation data means determining the evaluation result of the risk status of the media information in the corresponding dimension according to the evaluation data corresponding to the media information and the dimension. The higher the risk of the evaluation result of the risk status of the media information in the corresponding dimension, the greater the necessity of reviewing the corresponding media information from the perspective of the evaluation of the dimension. Still taking the media information as an example of advertisement, it is determined whether the corresponding advertisement slot is a widely disseminated advertisement slot according to the exposure data of the advertisement in the advertisement slot, and a score representing the breadth of the advertisement slot is obtained as the evaluation result of the corresponding advertisement in the advertisement slot evaluation dimension; it is determined whether the click-through rate of the corresponding advertisement is abnormal according to the click-through data of the advertisement in two adjacent time periods, and a score representing the degree of abnormality of the click-through rate of the advertisement is obtained as the evaluation result of the corresponding advertisement in the conversion evaluation dimension; it is determined whether the visits to the advertisement carrier are abnormal according to the visit data of the advertisement carrier in two adjacent time periods, and a score representing the degree of abnormality of the visits to the advertisement carrier is obtained as the evaluation result of the corresponding advertisement in the advertisement carrier evaluation dimension; it is determined whether the user who clicks the advertisement belongs to the black seed user group according to the object attribute data for determining whether the user who clicks the advertisement belongs to the black seed user group, and a score representing the index of the number of black seed users among the users who click the advertisement is obtained as the evaluation result of the corresponding advertisement in the user evaluation dimension; it is determined whether the industry to which the advertisement belongs is the specified industry according to the industry attribute data for determining whether the advertisement content belongs to the specified industry, and a weight score representing that the industry to which the advertisement belongs is an industry that needs to be reviewed is obtained as the evaluation result of the corresponding advertisement in the industry evaluation dimension.
[0034] Step 103: Determine the risk type of the media information in combination with the evaluation results corresponding to the dimensions.
[0035] Here, the risk type of media information can be divided into multiple different levels according to actual conditions. For example, the risk type of media information can be divided into two levels: high risk and low risk. High risk means that it is necessary to review whether the corresponding media information contains illegal content, and low risk means that it is not necessary to review whether the corresponding media information contains illegal content. It should be noted that the level of the risk type of media information is not limited to the two levels mentioned above. For example, the risk type of media information can also be divided into three levels: high risk, medium risk and low risk. High risk means that it is necessary to review whether the corresponding media information contains illegal content, medium risk means that it is necessary to review whether the corresponding media information contains illegal content, and low risk means that it is not necessary to review whether the corresponding media information contains illegal content.
[0036] By setting multiple dimensions for evaluating whether it is necessary to review the illegal content contained in the corresponding media information, the risk type of the media information is determined to be high risk or low risk according to the results of the risk status of the corresponding media information in the corresponding dimensions, that is, according to the evaluation results of the corresponding media information in the corresponding dimensions, the greater the necessity of review, so that by setting evaluation strategies in combination with different dimensions, it is convenient to comprehensively and objectively determine all media information with higher risks that needs to be reviewed to see whether it contains illegal content.
[0037] Step 105: When the media information is of a set risk type, the information delivery system is triggered to stop delivering the media information.
[0038] The information delivery system stops delivering the media information, which means that the information delivery system temporarily or permanently takes the media information offline in real time according to the result of determining the risk type of the media information. As an optional embodiment, the risk type can be set to different levels. When the media information is confirmed to be at the highest level, the information delivery system is triggered to stop the delivery plan of the media information and take the media information offline permanently; when the media information is confirmed to be at the second highest level, the information delivery system is triggered to temporarily stop the delivery plan of the media information and take the media information offline temporarily, and then further determine whether to resume the delivery of the media information according to the delivery plan.
[0039] As another optional embodiment, when it is determined that the media information is of a set risk type, the information delivery system is triggered to stop the delivery plan of the media information, temporarily offline the media information, and trigger a secondary review of the media information. Triggering a secondary review of the media information means recommending the media information determined to be of a set risk type to an audit system for secondary review or sending it to a user end for secondary review, and reconfirming the risk type of the recommended media information by triggering the secondary review process. As an optional embodiment, the media information recommendation is sent to the audit system for secondary review. When the media information is confirmed to be of a set risk type according to the secondary review, the media information is permanently offline, or when the media information is confirmed not to be of a set risk type according to the secondary review, the information delivery system is triggered to restore the media information online and continue to complete the delivery plan. Here, the secondary review refers to a review of whether the corresponding media information contains illegal content to determine the conclusion, and the secondary review is usually a manual review. By manually conducting a secondary review, the accuracy of the review conclusion can be ensured, and based on the results of the secondary review, when it is determined that the corresponding media information does contain illegal content, the corresponding media information will be removed from the information delivery system in a timely manner to avoid the adverse effects caused by the dissemination of the corresponding media information and the adverse effects on the reputation of the information delivery system.
[0040] In the above embodiment of the present invention, by setting multiple dimensions of media information, obtaining the evaluation data of the corresponding dimensions to determine the corresponding evaluation results, and combining the evaluation results of each dimension to determine the risk type of the media information, it is possible to efficiently and accurately screen out the media information that may contain illegal content according to the set dimensions, and automatically recommend these media information that may contain illegal content to the review system for secondary review to trigger the secondary review process. Through the secondary review, whether the media information contains illegal content is reconfirmed again. In this way, the media information of the set risk type can be comprehensively and objectively determined according to each dimension, ensuring the coverage of the evaluation of the risk type of the media information, so as to improve the review efficiency and review effect. Taking the media information as an advertisement as an example, the recommendation-based inspection system of the information inspection method of the embodiment of the present invention obtains the evaluation data of the corresponding dimensions of the advertisement currently delivered by the advertising delivery platform in real time, combines the evaluation results of each dimension to determine the risk of the advertisement, automatically recommends risky advertisements that may contain illegal content in real time and triggers secondary review accordingly, thereby achieving the inspection purpose of automatically recommending risky advertisements for manual review.
[0041] Optionally, when the media information is determined to be of a set risk type based on the evaluation results of each dimension of the media information, the media information is recommended for secondary review. In this way, on the one hand, high-risk media information that may contain illegal content can be comprehensively and objectively determined for secondary review based on multiple dimensions, which can ensure the quality of the review under the premise of reducing the workload of the secondary review; on the other hand, the risk type of the media information can be automatically identified and determined and recommended for secondary review, which can improve the efficiency of the review. Taking the media information as an advertisement as an example, by obtaining the evaluation data of the advertisements placed in the information delivery system in the corresponding dimensions, determining the evaluation results corresponding to each dimension, and determining the risk type of the advertisement, the high-risk advertisements that meet the conditions for the necessity of random inspection of whether they contain illegal content can be recommended for secondary review. In this way, high-risk advertisements that may contain illegal content can be comprehensively and objectively determined for secondary review based on multiple dimensions, which can ensure the quality of the advertising inspection under the premise of reducing the workload of the secondary review; secondly, high-risk advertisements can be automatically identified and determined and recommended for secondary review, which can improve the efficiency of the advertising inspection.
[0042] In some embodiments, before obtaining evaluation data of at least one dimension of the delivered media information from the information delivery system, the process includes:
[0043] Determine corresponding dimensions according to different attributes of the media information;
[0044] The attributes include at least one of the following: content attributes, carrier attributes, location attributes, conversion attributes and object attributes.
[0045] Here, the dimension refers to the evaluation angle used to evaluate whether it is necessary to further review the illegal content contained in the corresponding media information. Each dimension corresponds to an evaluation strategy, and each evaluation strategy is used to independently evaluate the degree of necessity of reviewing the illegal content contained in the media information. Among them, the dimension for evaluating whether it is necessary to further review the illegal content contained in the corresponding media information is determined according to different attributes of the media information, that is, the dimension corresponds to different attributes of the media information.
[0046] For example, the attributes of media information may include content attributes, and the dimension corresponding to the content attributes may refer to an industry evaluation dimension for determining whether the industry to which the media information belongs is a designated industry that requires key review based on the content of the media information; the attributes of media information may include carrier attributes, and the dimension corresponding to the carrier attributes may refer to an information display carrier evaluation dimension for determining whether the number of visits to the carrier displayed by the media information is abnormal based on the carrier on which the media information is displayed; the attributes of media information may include position attributes, and the dimension corresponding to the position attribute may refer to a display position evaluation dimension for determining whether the display position of the media information in the information delivery system is a display position that is very widely disseminated; the attributes of media information may include conversion attributes, and the dimension corresponding to the conversion attribute may refer to a conversion evaluation dimension for determining whether the conversion of the media information is abnormal based on the real-time conversion data of the media information; the attributes of media information may include object attributes, and the dimension corresponding to the object attribute may refer to a user evaluation dimension for determining whether the media information is abnormal based on whether the users who click on the media information include black seed users.
[0047] In the above-mentioned embodiment of the present invention, by formulating evaluation strategies of corresponding dimensions according to different attributes of media information, it is determined whether there is a need for further review of illegal content contained in the media information, so that an objective evaluation result of whether the media information is high-risk can be obtained through the objectivity of the different attributes of the media information. It can also ensure that media information with a high degree of necessity for further review can be comprehensively determined from different angles to ensure the quality of the inspection and audit.
[0048] In some embodiments, when the dimension includes a display position evaluation dimension, obtaining evaluation data of at least one dimension of the delivered media information from the information delivery system, and determining an evaluation result of the media information corresponding to the at least one dimension according to the evaluation data, includes:
[0049] The exposure number of the media information displayed in the corresponding information display position in the current delivery cycle in the information delivery system is obtained, and the score value corresponding to the display position evaluation dimension of the media information is determined according to the exposure number, wherein the score value corresponding to the display position evaluation dimension of the media information is obtained by taking the logarithm of the exposure number.
[0050] Here, the necessity of further reviewing the illegal content contained in the media information is determined from the display position evaluation dimension, mainly by evaluating the breadth of the dissemination range of the corresponding information display position of the media information. Among them, the evaluation data corresponding to the display position evaluation dimension includes the number of exposures of the media information on the corresponding display position, and the score value representing the breadth of the dissemination range of the information display position can be determined as shown in the following formula 1:
[0051] f i=log(pv) (Formula 1)
[0052] Among them, pv refers to the number of exposures of media information. When a user browses the display page of media information in the information delivery system, the display of media information on the display page is considered an exposure. i Refers to the score of media information in the display position evaluation dimension. The score of media information in the display position evaluation dimension f i It is a monotonically increasing function of the exposure number pv, and log represents the logarithm.
[0053] In the above embodiment of the present invention, by evaluating the necessity of further review of the media information from the display position evaluation dimension, for information display positions with a very wide range of dissemination, taking the media information as an advertisement as an example, for the head advertisement position with a very wide range of dissemination, once media information containing illegal content appears, the harm caused is also relatively large. In this way, by evaluating the risk type of the media information from the information display position evaluation dimension, it is possible to avoid missing the media information that needs further review.
[0054] In some embodiments, when the dimension includes a conversion evaluation dimension, obtaining evaluation data of at least one dimension of the delivered media information from the information delivery system, and determining an evaluation result of the media information corresponding to the at least one dimension according to the evaluation data includes:
[0055] The click rate of the media information delivered in the current delivery cycle in the information delivery system is obtained, and the score value corresponding to the media information in the conversion evaluation dimension is determined according to the increment of the click rate of the media information in the current sampling period of the statistical period, wherein when the increment is greater than 0, the score value corresponding to the media information in the conversion evaluation dimension is proportional to the increment, and when the increment is less than 0, the score value corresponding to the media information in the conversion evaluation dimension is zero.
[0056] Here, the necessity of further reviewing the illegal content contained in the media information is determined from the conversion evaluation dimension, mainly by evaluating the changes in the click data of the media information in the current sampling period. The statistical period of different media information can be determined according to the life cycle of the media information. The life cycle of the media information is usually the same as the display time of the media information in the information delivery system, and can be divided into multiple sampling periods according to the statistical period of the media information. Determining the changes in the click data of the media information in the current sampling period may refer to determining the difference between the click volume of the media information in the current sampling period and the click volume in the previous sampling period, or to determining the difference between the click volume of the media information in the current sampling period and the average click volume in its statistical period. As an optional embodiment, the evaluation data corresponding to the conversion evaluation dimension includes the click rate of the media information in two adjacent sampling periods within the statistical period, and the determination of the score value representing the abnormality of the click rate of the media information can be shown in the following formula 2:
[0057]
[0058] Among them, ctr now Refers to the click rate of media information in the current sampling period, ctr avg It refers to the average click rate of media information during the statistical period. i Refers to the score of the media information in the display position evaluation dimension. The click rate in the current sampling period is greater than the average click rate ctr now >ctr avg When, that is, the click rate increment △ = (ctr now -ctr avg ) / ctr avg When it is greater than 0, the larger the increment, the higher the score value representing the abnormality of the click rate of the media information, and the click rate in the current sampling period is less than the average click rate ctr now <ctr avg , that is, when the increment is less than 0, the score value representing the abnormality of the click rate of the media information is 0.
[0059] In the above-mentioned embodiment of the present invention, the click-through rate of media information is usually strongly correlated with the content contained in the media information, and media information containing illegal content usually tends to be produced through picture materials and texts, such as tending to use vulgar pictures, mouse hands, false software icons, etc., to induce users to click, so as to enhance its diffusion effect. Here, by evaluating the necessity of further review of media information from the conversion evaluation dimension, the media information with abnormal jitter in the click-through rate of the information display position is regarded as the focus of further review. In this way, by evaluating the risk type of media information from the conversion evaluation dimension, it is possible to avoid missing media information that needs to be stopped or further reviewed.
[0060] In some embodiments, when the dimension includes an information display carrier evaluation dimension, obtaining evaluation data of at least one dimension of the delivered media information from the information delivery system, and determining an evaluation result of the media information corresponding to the at least one dimension according to the evaluation data, includes:
[0061] Obtain the number of visits to the corresponding carrier of the media information delivered by the information delivery system in the current delivery cycle, and determine the score value corresponding to the information display carrier evaluation dimension of the media information based on the change in the number of visits to the corresponding carrier of the media information, wherein the score value corresponding to the information display carrier evaluation dimension of the media information is proportional to the number of visits in the current time period.
[0062] Here, the necessity of further reviewing the illegal content contained in the media information is determined from the information display carrier evaluation dimension, mainly by evaluating the changes in the number of visits to the carrier corresponding to the media information. Determining the changes in the number of visits to the carrier corresponding to the media information may refer to obtaining the number of visits to the corresponding carrier of the media information delivered by the information delivery system in the current delivery cycle, and determining the difference between the number of visits to the corresponding carrier of the media information in the current time period and the number of visits in the previous time period. As an optional embodiment, the evaluation data corresponding to the information display carrier evaluation dimension includes the number of visits to the carrier corresponding to the media information in different time periods, and the determination of the score value that characterizes the changes in the number of visits to the carrier corresponding to the media information in different time periods can be shown in the following formula three:
[0063]
[0064] Among them, PV T+1 Refers to the number of visits to the corresponding carrier of the media information within the time period of T+1, PV T It refers to the number of visits to the corresponding carrier of the media information within the time period T, f iRefers to the score of media information in the evaluation dimension of information display carrier. The number of visits in the current time period is greater than the number of visits in the previous time period. T+1 >PV T PV of the current time period T+1 The larger the value, or the visit volume increment △=(1+(PV T+1 -PV T ) / PV T The larger the value is, the higher the score value representing the abnormality of the number of visits to the corresponding carrier of the media information is. The number of visits in the current time period is less than the number of visits PV in the previous time period. T+1 <PV T , then the number of visits PV in the current time period T+1 The larger the value, the higher the score representing the abnormality of the access volume of the corresponding carrier of the media information.
[0065] In the above-mentioned embodiments of the present invention, the number of visits to the carrier corresponding to the media information is usually strongly correlated with the content contained in the media information. For some media information carriers with a wide user audience, there will be a large amount of media information inside. In the era of self-media, the spread of popular articles may reach tens of millions of visits in a short period of time. For the carriers corresponding to the media information with a large number of visits, when there is an abnormal change in the number of visits, if there is media information containing illegal content, the impact will be very bad. By evaluating the necessity of further review of the media information from the information display carrier evaluation dimension, the media information with abnormal changes in the number of visits to the corresponding carrier of the media information will be regarded as the focus of further review. In this way, by evaluating the risk type of the media information from the information display carrier evaluation dimension, it is possible to avoid missing the media information that needs to be stopped or further reviewed.
[0066] In some embodiments, when the dimension includes a user evaluation dimension, obtaining evaluation data of at least one dimension of the delivered media information from the information delivery system, and determining an evaluation result of the media information corresponding to the at least one dimension according to the evaluation data, includes:
[0067] Obtain the receiving user information of the media information delivered by the information delivery system in the current delivery cycle, determine the number of users of a specified type based on the receiving user information, and determine the score value corresponding to the media information in the user evaluation dimension, wherein the score value corresponding to the media information in the user evaluation dimension is the accumulated value of the users of the specified type.
[0068] Here, the necessity of further reviewing the illegal content contained in the media information is determined from the user evaluation dimension, mainly by evaluating whether the specified type of users are included in the receiving users of the media information. Among them, the receiving users of the media information refer to users who obtain the media information by clicking, browsing, downloading and other behaviors of receiving the media information. The specified type of user may refer to a black seed user that characterizes the user as a problem user. For example, the server can judge the user type based on the historical user behavior records, and divide the users according to the number and / or probability of each user's historical clicks on the media information containing illegal content, and form a black seed user group composed of problem users. Among them, determining the number of specified type of users based on the receiving user information of the media information may refer to determining whether it is a black seed user and counting the number of black seed users based on the object attributes of the user who clicked the media information. As an optional embodiment, the evaluation data corresponding to the user evaluation dimension includes the number of black seed users among the receiving users of the media information, and the determination of the score value characterizing whether the receiving users of the media information contain the specified type of users can be shown in the following formula four:
[0069] f i =∑g(user k )*τ(user k ∈BlackSeed)) (Formula 4)
[0070] Among them, BlackSeed represents the black seed user. If user k Belongs to black seed users, τ(user k ∈BlackSeed)=1, otherwise τ(user k ∈BlackSeed)=0,(user k )∈[0,1] represents the score of whether the corresponding receiving user of the media information is a black seed user, f i It refers to the score value of the media information corresponding to the user evaluation dimension. The score value of the media information corresponding to the user evaluation dimension is f i is the accumulated value of the number of black seed users.
[0071] In the above embodiment of the present invention, the quality of the corresponding receiving user of the media information is strongly related to the content contained in the media information. The users are classified according to the behavior of the users who receive the media information, and the problematic users who spread illegal content or like to spread illegal content are recorded. When the corresponding receiving users include a large number of problematic users' media information, the possibility of containing illegal content is also very high. By evaluating the necessity of further review of the media information from the user evaluation dimension, the media information whose receiving users include a large number of problematic users is regarded as the focus of further review. In this way, by evaluating the risk type of the media information from the user evaluation dimension, it is possible to avoid missing the media information that needs to be stopped or further reviewed.
[0072] In some embodiments, when the dimension includes an industry evaluation dimension, obtaining evaluation data of at least one dimension of the delivered media information from the information delivery system, and determining the evaluation result of the media information corresponding to the at least one dimension according to the evaluation data, includes:
[0073] Obtain the industry attributes of the media information delivered by the information delivery system in the current delivery cycle, determine the industry and industry weight corresponding to the media information based on the industry attribute information, and determine the score value corresponding to the evaluation dimension of the industry to which the media information belongs, wherein, when the industry corresponding to the media information belongs to a specified industry, the score value corresponding to the evaluation dimension of the industry to which the media information belongs is the accumulated value of the corresponding industry weight, and when the industry corresponding to the media information does not belong to the specified industry, the score value corresponding to the evaluation dimension of the industry to which the media information belongs is zero.
[0074] Here, the necessity of further reviewing the illegal content contained in the media information is determined from the industry assessment dimension, mainly by determining whether the industry to which the media information belongs is a designated industry that needs to be strictly controlled. Among them, whether the designated industry that needs to be strictly controlled refers to the industry determined according to laws, regulations, policy adjustments, etc. Secondly, according to the degree to which the industry needs to be reviewed, different weights can be set for different designated industries. For example, some industries that need to be reviewed in particular can be set with higher weight values to ensure that media information belonging to the industry can be selected and recommended for further review. Among them, the industry set of designated industries and the weights corresponding to each industry can be predetermined. Determining the corresponding industry and industry weight based on the industry information to which the media information belongs can refer to determining whether it is a designated industry and obtaining the corresponding weight of the designated industry based on the industry attributes of the media information. As an optional embodiment, the evaluation data corresponding to the industry evaluation dimension includes the industry attributes corresponding to the media information, and the determination of the score value representing whether the industry to which the media information belongs is a designated industry that needs to be strictly controlled can be shown in the following formula five:
[0075] f i =∑K i *τ(ad∈KeyIndustry) (Formula 5)
[0076] Among them, KeyIndustry represents the industry set of the specified industry that needs to be strictly checked. If τ(ad∈keyIndustry)=1, it means that the advertisement belongs to the specified industry that needs to be strictly checked. If τ(ad∈keyIndustry)=0, it means that the advertisement does not belong to the specified industry that needs to be strictly checked. i represents the weight of the corresponding industry, f i Refers to the score value of the media information in the industry evaluation dimension. When the industry corresponding to the media information belongs to the specified industry, the score value of the media information in the industry evaluation dimension is f i is the industry weight value K of the corresponding industry i The cumulative value of .
[0077] In the above embodiment of the present invention, the industry to which the media information belongs is strongly related to the content contained in the media information. By pre-establishing an industry set of designated industries that need to be strictly checked, when the media information is determined to belong to an industry included in the industry set, the possibility that it contains illegal content is very high. Among them, by evaluating the necessity of further review of the media information from the industry assessment dimension, the media information that belongs to the designated industry that needs to be strictly checked and reviewed is regarded as a key focus object that needs further review. In this way, by evaluating the risk type of the media information from the industry assessment dimension, it is possible to avoid missing the media information that needs to be stopped or further reviewed.
[0078] In some embodiments, combining the evaluation results corresponding to the dimensions to determine the risk type of the media information includes:
[0079] The score values of the dimensions are weighted according to corresponding weight parameters to determine a final score value of the media information, and the risk type of the media information is determined according to the final score value.
[0080] The dimensions may include at least two of the following: industry evaluation dimension, information display carrier evaluation dimension, display position evaluation dimension, conversion evaluation dimension, and user evaluation dimension. Here, multiple dimensions are predetermined to evaluate whether it is necessary to review the illegal content contained in the corresponding media information, and corresponding weight parameters are set for different dimensions. The score values corresponding to the media information in different dimensions are weighted and summed based on the corresponding weight parameters to obtain the final score value used to characterize whether the media information is a high-risk type of media information that requires further secondary review.
[0081] The industry assessment dimension refers to the assessment dimension of determining whether the industry to which the media information belongs is a designated industry that requires key review based on its content; the information display carrier assessment dimension refers to the assessment dimension of determining whether the traffic of the carrier displayed by the media information is abnormal, so as to determine whether it is necessary to stop the delivery of the media information and / or recommend it for a secondary review; the display position assessment dimension refers to the assessment dimension of determining whether the display position of the media information in the information delivery system is a very widely disseminated display position, so as to determine whether it is necessary to stop the delivery of the media information and / or recommend it for a secondary review; the conversion assessment dimension refers to the assessment dimension of determining whether the conversion of the media information is abnormal based on the real-time conversion data of the media information, so as to determine whether it is necessary to stop the delivery of the media information and / or recommend it for a secondary review; the user assessment dimension refers to the assessment dimension of determining whether the media information is abnormal based on whether the users who click on the media information include black seed users, so as to determine whether it is necessary to stop the delivery of the media information and / or recommend it for a secondary review.
[0082] The weight parameters corresponding to different dimensions can be determined by training a linear weighted model, or by training a model through deep learning, or by empirical values. Figure 5 In an optional embodiment of the present invention, each dimension includes an industry evaluation dimension, an information display carrier evaluation dimension, a display position evaluation dimension, a conversion evaluation dimension, and a user evaluation dimension. The weight parameters corresponding to each dimension can be as follows: Figure 5 As listed in.
[0083] In some embodiments, before weighting the score values corresponding to the dimensions according to the corresponding weight parameters, the process includes:
[0084] Constructing a training data set, wherein the training data set includes evaluation results and corresponding risk types corresponding to the training media information in each of the dimensions;
[0085] The linear model is trained based on the training data set until the corresponding loss function converges, thereby obtaining weight parameters in the linear model corresponding to each of the dimensions.
[0086] Here, a linear model that linearly weights the score values of each dimension is pre-established, and a training data set is constructed to train the linear model to determine the weight parameters corresponding to each dimension. The training data set can be composed of media information that is known to need to stop delivery and / or need to undergo a secondary review as training media information. The linear model is trained by forming a training data set by annotating the training media information and its corresponding risk type. For media information that is determined to need to stop delivery and / or undergo a secondary review, the corresponding output of the training model is 1, otherwise the output is 0. In this way, the model parameters in the training model are adjusted by back-propagating the error of the training output result, and the training is continuously iterated until the loss function converges to obtain the final linear model.
[0087] Among them, the prediction function of the linear model can be shown as follows:
[0088]
[0089] The loss function corresponding to the linear model can adopt the stochastic gradient descent (SGD) function, and the loss function can be shown in the following formula 7:
[0090] logloss=∑y i ln(p i )+(1-y i )ln(1-p i ) (Formula 7)
[0091] Here, for the training media information that is determined to require secondary review, y i = 1, for training media information that is determined not to require secondary review, y i =0.
[0092] In some embodiments, combining the evaluation results corresponding to the dimensions to determine the risk type of the media information includes:
[0093] Constructing input features corresponding to the dimensions according to the evaluation results of the dimensions;
[0094] The input features and the image data of the media information are respectively used as inputs of a combined model of a trained linear model and a neural network, and the risk type of the media information is determined according to the output result after the first output of the linear model and the second output of the neural network are input into a logistic regression layer by the combined model.
[0095] Here, a combination model of a linear model and a neural network is established by introducing deep learning to perform training, so as to determine the weight parameters corresponding to each of the dimensions. The data for training the combination model of the linear model and the neural network includes input features corresponding to the dimensions respectively constructed according to the evaluation results of each of the dimensions, and the image data of the media information. The input features of the media information constructed according to the evaluation results of each of the dimensions are used as the input of the linear model in the combination model, the image data of the media information is used as the input of the neural network in the combination model, and the first output of the linear model and the second output of the neural network are used as the input to the logistic regression layer, so as to determine the risk type of the media information according to the 0-1 probability result output by the logistic regression layer.
[0096] In some embodiments, before using the input features and the image data of the media information as inputs of the trained linear model and the combined model of the neural network, the method further comprises:
[0097] Constructing a sample data set, the sample data set including sample input features constructed from the evaluation results of the sample media information corresponding to the set dimension, sample image data of the sample media information, and corresponding risk types;
[0098] Construct an initial linear model and a combined model of a neural network, use the sample input features of the sample media information as the input of the linear model, use the sample image data and the corresponding risk type as the input of the neural network, use the first training output of the linear model and the second training output of the neural network as the input of the logistic regression layer, adjust the network parameters in the initial combined model according to the error between the output of the logistic regression layer and the risk type of the corresponding sample media information, and perform iterative training until the corresponding loss function converges.
[0099] Here, by pre-establishing an initial linear model and a combined model of a neural network, a sample data set is constructed to train the combined model to determine the weight parameters corresponding to each of the dimensions. The acquisition of the sample data set may include: taking media information that needs to be stopped and / or needs to be reviewed again as sample media information; constructing sample input features according to the evaluation results corresponding to the set dimensions of the sample media information; collecting images corresponding to the sample media information as sample image data; and labeling the risk type of the sample image data.
[0100] The process of training the initial combined model includes: taking the sample input features of the sample media information as the input of the linear model, taking the sample image data and the corresponding risk type annotation as the input of the neural network, taking the first training output of the linear model and the second training output of the neural network as the input of the logistic regression layer, and for the sample media information that needs to be stopped and / or undergo a secondary review, the corresponding output of the logistic regression layer is 1, otherwise the output is 0, and the risk type problem of the media information is converted into a 0-1 classification problem through the output of the logistic regression layer, and the output error of the logistic regression layer is back-propagated in the combined model. In each network layer of the back-propagation, various gradient solving methods are used to determine the gradient of the loss function relative to the parameters of each network layer, and the parameters of the network layer are updated by subtracting the corresponding gradient. The network parameters in the initial combined model are adjusted by continuous training and iteration until the loss function converges to obtain the trained combined model, thereby obtaining the weight parameters corresponding to each dimension.
[0101] See also Figure 6, is a schematic diagram of the architecture of a combination model of a linear model and a neural network provided in an optional embodiment of the present invention, and the combination model is also called a wide and deep model (Wide & Deep Models), wherein the wide model (WideModels) corresponds to the linear model, and the input of the Wide Models includes input features constructed based on the evaluation results corresponding to each of the dimensions, such as: input feature X1 constructed according to the evaluation result corresponding to the industry evaluation dimension, input feature X2 constructed according to the evaluation result corresponding to the information display carrier evaluation dimension, input feature X3 constructed according to the evaluation result corresponding to the display position evaluation dimension, input feature X4 constructed according to the evaluation result corresponding to the conversion evaluation dimension, and input feature X5 constructed according to the evaluation result corresponding to the user evaluation dimension. Deep Models correspond to neural networks. Deep Models include input layers, encoding layers, one or more hidden layers, and output layers. The input of Deep Models includes sample image data and corresponding risk type labels. The input layer extracts sparse features of sample image data. The encoding layer encodes the extracted features and maps them to a relatively lower-dimensional, dense encoding space. This process is also called dense embeddings, which can solve the problem of feature sparsity. The hidden layer can fit the encoded values output by the encoding layer to the same value space to obtain a lower-dimensional feature vector. The feature pairs of the last layer of Wide Models and Deep Models are uniformly input to the logistic regression layer. The logistic regression layer uses an activation function such as the sigmoid function, which can be shown in the following formula 8:
[0102]
[0103] Where Y represents the label of the binary classification, σ(·) represents the sigmoid function, φ(x) represents the vector dot product transformation (cross product transformations) of the original feature x, and b represents the bias term. wide Represents the weight vector of Wide Models, W deep Indicates the application of the final activation function a (lf) The weight on .
[0104] In the above embodiment of the present invention, by introducing a deep learning method to perform modeling on the basis of linear weighting, a combined model of a linear model and a neural network is constructed. Not only can deterministically observable feature data be modeled, such as modeling input features constructed based on evaluation results corresponding to each dimension of the media information, but also sparse features or behavioral data that cannot be directly deterministically observable can be modeled, such as modeling features carried in image data of media information, thereby forming a wide deep learning framework that realizes memory and generalization in one model, which can more comprehensively and accurately identify media information of a set risk type, and stop the delivery of media information containing illegal content, and / or recommend media information that needs further review to determine whether it contains illegal content for a secondary review.
[0105] In some embodiments, the information inspection method provided by the embodiment of the present invention further includes:
[0106] Acquire feedback information, and when determining according to the feedback information that the media information is of a set risk type, trigger the information delivery system to stop delivering the media information.
[0107] The feedback information includes information obtained from the designated user feedback portal reflecting whether the corresponding media information contains illegal content. Information delivery systems used to display media information usually have user feedback portals, which facilitate users to make suggestions to the information delivery system through the user feedback portal, including feedback on whether the media information displayed in the information delivery system contains illegal content. Please refer to Figure 7 Taking WeChat, an information delivery system, as an example, a user feedback portal is provided at the top of the advertisement page, and users can provide feedback on possible problems with the corresponding advertisement by selecting the user feedback portal.
[0108] In the above-mentioned embodiment of the present invention, by obtaining feedback information on whether the media information contains illegal content, determining the risk type of the media information based on the feedback information is used as an auxiliary means, so that the media information that needs to be stopped from being delivered can be more comprehensively and accurately identified, and / or the media information that needs to be further reviewed to determine whether it contains illegal content can be more comprehensively and accurately identified and recommended to the review system for secondary review.
[0109] In some embodiments, when determining according to the feedback information that the media information is of a set risk type, the step further includes: recommending the media information for secondary review.
[0110] Here, recommending the media information for secondary review means: recommending the media information to the review system for manual review. The review system refers to a system that is relatively independent of the recommendation-based inspection system. Please refer to Figure 1 and Figure 2 The implementation side of the audit system can be a user terminal. Through the audit system loaded on the user terminal, it is convenient for the auditor to directly operate the user terminal to perform manual secondary audit on the media information recommended by the recommendation-based inspection system. The user terminal can be a computer device terminal such as a personal computer, a mobile terminal, etc.
[0111] In some embodiments, when the media information is of a set risk type, the information delivery system is triggered to stop delivering the media information, including: when the media information is of a set risk type, recommending the media information for secondary review and triggering the information delivery system to suspend delivery of the media information; based on the result of the secondary review confirming that the media information belongs to the set risk type, permanently removing the media information from the information delivery system.
[0112] Among them, the information delivery system stops delivering the media information, which may include stopping delivering the media information, temporarily removing the media information from the system delivery system, and recommending the media information for secondary review; or stopping delivering the media information and permanently removing the media information from the information delivery system. Secondary review refers to a review to determine whether the corresponding media information contains any illegal content, and the secondary review is usually a manual review. Through manual secondary review, the accuracy of the review conclusion can be ensured, so that according to the results of the secondary review, when it is determined that the corresponding media information does contain illegal content, the corresponding media information will be removed from the information delivery system in a timely and permanent manner to avoid the adverse effects caused by the spread of the corresponding media information and the adverse effects on the reputation of the information delivery system.
[0113] In some embodiments, the information inspection method further includes: when the media information is of a set risk type, displaying the media information in sequence according to the evaluation results, and displaying the evaluation results corresponding to each dimension of the media information.
[0114] Here, displaying the media information in order according to the evaluation results may refer to combining the media information to be recommended with the final score values obtained from the evaluation results corresponding to each dimension, displaying the media information in order according to the score values, and displaying the evaluation results corresponding to each dimension of the media information. Figure 8, which is a schematic diagram of displaying the media information in sequence according to the evaluation results, and displaying the media information determined as the set risk type in the inspection display interface, and the display content of each media information includes the content of the media information, such as text, image, and / or video information, and the evaluation results of the media information in the corresponding dimensions, such as the corresponding score values in the industry evaluation dimension, information display carrier evaluation dimension, display position evaluation dimension, conversion evaluation dimension and user evaluation dimension. Optionally, the display content of each media information can also include the weight value of the media information in the corresponding dimensions.
[0115] In the above-mentioned embodiment of the present invention, by displaying the media information in sequence according to the evaluation results in the inspection display interface, it is convenient for the recommendation-based inspection system that implements the information inspection method to intuitively display the media information recommended to the review system. Through the display interface, the evaluation results of the media information recommended for secondary review in different dimensions can be quickly understood, and it can be judged whether there are any abnormalities in the system; the different dimensions of the recommended media information can also be analyzed in combination with the experience value, and the evaluation results or the weights of different dimensions can be further adjusted and optimized.
[0116] In order to more clearly understand the implementation process of the information inspection method provided in the embodiment of the present invention, please refer to Fig. 9 The following is an optional specific example to illustrate the process of the information inspection method. The media information refers to an advertisement. The method includes the following steps:
[0117] S21, establishing a recommendation-based inspection model; wherein, the recommendation-based inspection model may be a linear weighted model, the prediction function of the linear model may be as shown in the aforementioned formulas 6 and 7, or a combination model of a linear model and a neural network that introduces deep learning, the prediction function of the combination model may be as shown in the aforementioned formula 8.
[0118] S22, construct a training data set, and train the recommendation-based inspection model through the training data set until the loss function of the model converges; the training data set includes advertising data known to contain illegal content and its corresponding risk type, and the scoring values of the advertising data in each dimension. Among them, the risk type of the advertisement known to contain illegal content is high-risk advertisement, and the risk type of the advertisement known not to contain illegal content is low-risk advertisement.
[0119] S23, obtaining evaluation data corresponding to the set dimensions of the advertisements currently delivered in the information delivery system, and determining the rating values according to the evaluation strategies of the corresponding dimensions; wherein the dimensions include the industry evaluation dimension, the information display carrier evaluation dimension, the display position evaluation dimension, the conversion evaluation dimension, and the user evaluation dimension. The corresponding rating value determined according to the evaluation strategy of the display position evaluation dimension can be as shown in the aforementioned formula one, the corresponding rating value determined according to the evaluation strategy of the conversion evaluation dimension can be as shown in the aforementioned formula two, the corresponding rating value determined according to the evaluation strategy of the information display carrier evaluation dimension can be as shown in the aforementioned formula three, the corresponding rating value determined according to the evaluation strategy of the user evaluation dimension can be as shown in the aforementioned formula four, and the corresponding rating value determined according to the evaluation strategy of the industry evaluation dimension can be as shown in the aforementioned formula five.
[0120] S24, weighting the score values corresponding to each dimension through the inspection model based on recommendation, and outputting a classification result for characterizing whether the corresponding advertisement is a high-risk advertisement; wherein, when the inspection model based on recommendation is a linear weighted model, the score values corresponding to the advertisement in each dimension are weighted based on the weight parameters corresponding to the trained model through the linear weighted model to obtain the final score of the advertisement, and the final score of the advertisement is used to characterize whether the corresponding advertisement is a classification result of a high-risk advertisement that requires further review. When the inspection model based on recommendation is a combination model of a linear model and a neural network, the score values corresponding to the advertisement in each dimension are weighted based on the weight parameters corresponding to the trained model through the combination model of the linear model and the neural network, and the features in the image data of the advertisement are extracted and identified, and a binary classification result is output that characterizes whether the corresponding advertisement is a high-risk advertisement that requires further review.
[0121] S25, recommending high-risk advertisements to the review system for secondary review according to the classification results output by the recommendation-based inspection model.
[0122] In the above-mentioned embodiment of the present invention, the inspection model based on recommendation combines the advertising position, advertising conversion, object attributes, etc., and makes real-time adjustments according to the policy. The inspection model based on recommendation is trained according to the historical audit data to optimize the model. The inspection model based on recommendation can recommend high-risk advertisements that may have the risk of violation in real time for secondary review, which can effectively reduce the number of advertisement inspections, and the inspections are more targeted, improving timeliness and reducing risks. In this way, the inspection model based on recommendation can be used to mine high-risk advertisements in online advertisements in real time and recommend them to the audit system, and the parameters of the inspection model based on recommendation can be adjusted in a personalized manner, such as setting the industry weight of the specified industry to respond to the policy in real time, ensuring that all advertisements in special specified industries must be selected to be recommended for secondary review, and according to the result of the secondary review confirming that the advertisement is a high-risk advertisement, the advertisement is permanently removed from the information delivery system to avoid the adverse effects caused by the spread of advertisements containing illegal content.
[0123] Another aspect of the present invention provides an embodiment in which the information inspection device can be implemented by a computer device such as a server. For the hardware structure of the information inspection device implementing the information inspection method, please refer to Fig.10 , is a schematic diagram of an optional hardware structure of an information inspection device provided in an embodiment of the present invention, including: at least one processor 901, a memory 902, at least one network interface 904 and a user interface 906. The various components in the information inspection device are coupled together through a bus system 905. It can be understood that the bus system 905 is used to realize the connection and communication between these components. In addition to the data bus, the bus system 905 also includes a power bus, a control bus and a status signal bus. For the sake of clarity, Fig. 9 In the specification, various buses are labeled as bus systems.
[0124] The user interface 906 may include a display, a keyboard, a mouse, a trackball, a click wheel, keys, buttons, a touch pad or a touch screen.
[0125] It is understood that the memory 902 can be a volatile memory or a non-volatile memory, and can also include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as static random access memory (SRAM), synchronous static random access memory (SSRAM). The memory described in the embodiments of the present invention is intended to include but is not limited to these and any other suitable categories of memory.
[0126] The memory 902 in the embodiment of the present invention is used to store various categories of data to support the operation of the information inspection device. Examples of these data include: any executable program used to operate on the information inspection device, such as an operating system and an application; corresponding evaluation data of the media information in each set dimension; the evaluation results corresponding to the media information in the dimension, etc.; wherein the operating system includes various system programs, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks. The application program may include various applications, such as a target application, a media player (Media Player), a browser (Browser), etc., for implementing various application services. The information inspection device that implements the information inspection method provided in the embodiment of the present invention may be included in the application program.
[0127] The method disclosed in the above embodiment of the present invention can be applied to the processor 901, or implemented by the processor 901. The processor 901 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit in the processor 901 or the instruction in the form of software. The above processor 901 may be a general processor, a digital signal processor (DSP, Digital Signal Processor), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The processor 901 can implement or execute the various methods, steps and logic block diagrams disclosed in the embodiment of the present invention. The general processor 901 can be a microprocessor or any conventional processor, etc. In combination with the implementation steps of the information inspection method provided in the embodiment of the present invention, it can be directly embodied as a hardware decoding processor to execute, or it can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium, which is located in a memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.
[0128] In an exemplary embodiment, the information inspection device can be used by one or more application specific integrated circuits (ASIC), DSP, programmable logic device (PLD), complex programmable logic device (CPLD) to execute the above method.
[0129] In the exemplary embodiment, please continue to see Fig.10 The information inspection device provided by an embodiment of the present invention includes: an evaluation module 11, a risk determination module 13 and a recommendation review module 15. The evaluation module 11 is used to obtain evaluation data of at least one dimension of the delivered media information from the information delivery system in real time, and determine the evaluation result of the media information corresponding to the at least one dimension according to the evaluation data; the risk determination module 13 is used to determine the risk type of the media information in combination with the evaluation results corresponding to each dimension; the stop delivery module 15 is used to trigger the information delivery system to stop delivering the media information when the media information is of a set risk type.
[0130] In some embodiments, the device further comprises a dimension determination module for determining corresponding dimensions according to different attributes of the media information; wherein the attributes comprise at least one of the following: content attributes, carrier attributes, location attributes, conversion attributes and object attributes.
[0131] In some embodiments, the evaluation module 11 is also used to obtain the exposure number of the media information displayed in the corresponding information display position in the current delivery cycle in the information delivery system when the dimension includes the display position evaluation dimension, and determine the score value corresponding to the display position evaluation dimension of the media information according to the exposure number, wherein the score value corresponding to the display position evaluation dimension of the media information is obtained by taking the logarithm of the exposure number.
[0132] In some embodiments, the evaluation module 11 is also used to obtain the click-through rate of the media information delivered in the information delivery system in the current delivery cycle when the dimension includes a conversion evaluation dimension, and determine the score value corresponding to the media information in the conversion evaluation dimension according to the increment of the click-through rate of the media information in the current sampling period of the statistical period, wherein when the increment is greater than 0, the score value corresponding to the media information in the conversion evaluation dimension is proportional to the increment, and when the increment is less than 0, the score value corresponding to the media information in the conversion evaluation dimension is zero.
[0133] In some embodiments, the evaluation module 11 is also used to obtain the number of visits to the carrier corresponding to the media information delivered by the information delivery system in the current delivery cycle when the dimension includes the information display carrier evaluation dimension, and determine the score value corresponding to the information display carrier evaluation dimension of the media information according to the change in the number of visits to the carrier corresponding to the media information, wherein the score value corresponding to the information display carrier evaluation dimension of the media information is proportional to the number of visits in the current time period.
[0134] In some embodiments, the evaluation module 11 is also used to obtain the receiving user information of the media information delivered by the information delivery system in the current delivery cycle when the dimension includes the user evaluation dimension, determine the number of users of a specified type based on the receiving user information, and determine the score value corresponding to the media information in the user evaluation dimension, wherein the score value corresponding to the media information in the user evaluation dimension is the accumulated value of the users of the specified type.
[0135] In some embodiments, the evaluation module 11 is also used to obtain the industry attributes of the media information delivered by the information delivery system in the current delivery cycle when the dimension includes an industry evaluation dimension, determine the industry and industry weight corresponding to the media information based on the industry attribute information, and determine the score value corresponding to the media information in the industry evaluation dimension, wherein, when the industry corresponding to the media information belongs to a specified industry, the score value corresponding to the media information in the industry evaluation dimension is the accumulated value of the corresponding industry weight, and when the industry corresponding to the media information does not belong to the specified industry, the score value corresponding to the media information in the industry evaluation dimension is zero.
[0136] In some embodiments, the risk determination module 13 is further used to weight the score values of each dimension according to corresponding weight parameters to determine a final score value of the media information, and determine the risk type of the media information according to the final score value.
[0137] In some embodiments, the risk determination module 13 is also used to construct a training data set, which includes the evaluation results and corresponding risk types of the training media information corresponding to each of the dimensions; the linear model is trained based on the training data set until the corresponding loss function converges, and the weight parameters corresponding to each of the dimensions in the linear model are obtained.
[0138] In some embodiments, the risk determination module 13 is further used to construct input features corresponding to the dimensions according to the evaluation results corresponding to the dimensions; the input features and the image data of the media information are respectively used as inputs of a combined model of a trained linear model and a neural network, and the risk type of the media information is determined according to the output result after the first output of the linear model and the second output of the neural network are input into the logistic regression layer according to the combined model.
[0139] In some embodiments, the risk determination module 13 is also used to construct a sample data set, which includes sample input features constructed from the evaluation results of the sample media information corresponding to each dimension, sample image data of the sample media information and the corresponding risk type; construct an initial linear model and a combined model of a neural network, use the sample input features of the sample media information as inputs to the linear model, use the sample image data and the corresponding risk type as inputs to the neural network, use the first training output of the linear model and the second training output of the neural network as inputs to a logistic regression layer, adjust the network parameters in the initial combined model according to the error between the output of the logistic regression layer and the risk type of the corresponding sample media information, and perform iterative training until the corresponding loss function converges.
[0140] In some embodiments, the stop delivery module 15 is used to recommend the media information for secondary review when the media information is of a set risk type; based on the result of the secondary review confirming that the media information belongs to the set risk type, the media information is permanently removed from the information delivery system.
[0141] In some embodiments, the device further includes a feedback processing module, which is used to obtain feedback information, and when it is determined according to the feedback information that the media information is of a set risk type, trigger the information delivery system to stop delivering the media information.
[0142] In some embodiments, the device further includes a display module for displaying the media information in sequence according to the evaluation results when the media information is of a set risk type, and displaying the evaluation results of the media information corresponding to each of the dimensions.
[0143] It should be noted that: the information inspection device provided in the above embodiment only uses the division of the above program modules as an example to illustrate when implementing the information inspection method in the embodiment of the present invention. In actual applications, the above processing can be assigned to different program modules as needed, that is, the internal structure of the device is divided into different program modules to complete all or part of the processing described above. In addition, the information inspection device provided in the above embodiment and the information inspection method embodiment in the embodiment of the present invention belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.
[0144] The embodiment of the present invention also provides a storage medium, such as Fig.10 The memory 902 shown includes an executable computer program, and the above computer program can be executed by the processor to complete the steps of the information inspection method provided in the embodiment of the present invention. The readable storage medium can refer to a magnetic random access memory (FRAM), a read-only memory (ROM), a programmable read-only memory (PROM), a non-volatile read-only memory (EPROM), an electrically pluggable programmable read-only memory (EEPROM), a flash memory, a magnetic surface memory, an optical disk, or a compact disk read-only memory (CD-ROM) and other memories; it can also be various devices including one or any combination of the above memories, such as a computer device.
[0145] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.
Claims
1. An information inspection method, characterized in that: include: Acquire evaluation data of at least one dimension of the delivered media information in real time from the information delivery system, When the dimension includes a conversion evaluation dimension, the click rate of the media information delivered in the information delivery system in the current delivery cycle is obtained, the difference between the click rate of the media information in the current sampling period of the statistical period and the average click rate in the statistical period is calculated, the ratio of the difference to the average click rate in the statistical period is used as the increment of the click rate in the current sampling period, and the score value corresponding to the media information in the conversion evaluation dimension is determined based on the increment, wherein when the increment is greater than 0, the score value corresponding to the media information in the conversion evaluation dimension is proportional to the increment, and when the increment is less than 0, the score value corresponding to the media information in the conversion evaluation dimension is zero; Constructing input features corresponding to the dimensions according to the evaluation results corresponding to the dimensions; The input features and the image data of the media information are respectively used as inputs of a combined model of a trained linear model and a neural network, and the risk type of the media information is determined according to an output result after the combined model inputs the first output of the linear model and the second output of the neural network into a logistic regression layer, wherein the evaluation result is characterized as a score value corresponding to the evaluation dimension; When the media information is of a set risk type, the information delivery system is triggered to stop delivering the media information.
2. The method according to claim 1, characterized in that Before obtaining evaluation data of at least one dimension of the delivered media information in real time from the information delivery system, the method includes: Determine corresponding dimensions according to different attributes of the media information; The attributes include at least one of the following: content attributes, carrier attributes, location attributes, conversion attributes and object attributes.
3. The method according to claim 1, characterized in that The method further comprises: When the dimension includes a display position evaluation dimension, the exposure number of the media information displayed in the corresponding information display position in the current delivery cycle in the information delivery system is obtained, and the score value corresponding to the display position evaluation dimension of the media information is determined according to the exposure number, wherein the score value corresponding to the display position evaluation dimension of the media information is obtained by taking the logarithm of the exposure number.
4. The method according to claim 1, characterized in that The method further comprises: When the dimension includes an information display carrier evaluation dimension, the number of visits to the carrier corresponding to the media information delivered by the information delivery system in the current delivery cycle is obtained, and the score value corresponding to the information display carrier evaluation dimension is determined based on the change in the number of visits to the carrier corresponding to the media information, wherein the score value corresponding to the information display carrier evaluation dimension is proportional to the number of visits in the current time period.
5. The method according to claim 1, characterized in that The method further comprises: When the dimension includes a user evaluation dimension, the information delivery system obtains information about users who receive the media information delivered during the current delivery cycle, determines the number of users of a specified type based on the received user information, and determines the score value corresponding to the media information in the user evaluation dimension, wherein the score value corresponding to the media information in the user evaluation dimension is the accumulated value of the users of the specified type.
6. The method according to claim 1, characterized in that The method further comprises: When the dimension includes an industry evaluation dimension, the industry attributes of the media information delivered by the information delivery system in the current delivery cycle are obtained, and the industry and industry weight corresponding to the media information are determined according to the industry attribute information, and the score value corresponding to the media information in the industry evaluation dimension is determined, wherein, when the industry corresponding to the media information belongs to a specified industry, the score value corresponding to the media information in the industry evaluation dimension is the accumulated value of the corresponding industry weight, and when the industry corresponding to the media information does not belong to the specified industry, the score value corresponding to the media information in the industry evaluation dimension is zero.
7. The method according to any one of claims 1 to 6, characterized in that The method further comprises: The score values of the dimensions are weighted according to corresponding weight parameters to determine a final score value of the media information, and the risk type of the media information is determined according to the final score value.
8. The method according to claim 7, characterized in that Before weighting the score values of each dimension according to the corresponding weight parameters, the method includes: Constructing a training data set, wherein the training data set includes evaluation results and corresponding risk types corresponding to the training media information in each of the dimensions; The linear model is trained based on the training data set until the corresponding loss function converges, thereby obtaining weight parameters in the linear model corresponding to each of the dimensions.
9. The method according to claim 1, characterized in that Before using the input features and the image data of the media information as inputs of the trained linear model and the combined model of the neural network, the method further comprises: Constructing a sample data set, the sample data set including sample input features constructed from evaluation results corresponding to the sample media information in each of the dimensions, sample image data of the sample media information, and corresponding risk types; Construct an initial linear model and a combined model of a neural network, use the sample input features of the sample media information as the input of the linear model, use the sample image data and the corresponding risk type as the input of the neural network, use the first training output of the linear model and the second training output of the neural network as the input of the logistic regression layer, adjust the network parameters in the initial combined model according to the error between the output of the logistic regression layer and the risk type of the corresponding sample media information, and perform iterative training until the corresponding loss function converges.
10. The method according to any one of claims 1 to 6, characterized in that When the media information is of a set risk type, triggering the information delivery system to stop delivering the media information specifically includes: When the media information is of a set risk type, the media information is recommended for secondary review and the information delivery system is triggered to suspend delivery of the media information; According to the result of the secondary review confirming that the media information belongs to the set risk type, the media information is permanently removed from the information delivery system.
11. The method according to any one of claims 1 to 6, characterized in that Also includes: When the media information is of a set risk type, the media information is displayed in sequence according to the evaluation results, and the evaluation results corresponding to the media information in each dimension are displayed.
12. An information inspection device, characterized in that: include: An evaluation module is used to obtain evaluation data of at least one dimension of the delivered media information from the information delivery system in real time, and when the dimension includes a conversion evaluation dimension, obtain the click rate of the media information delivered in the information delivery system in the current delivery cycle, calculate the difference between the click rate of the media information in the current sampling period of the statistical period and the average click rate in the statistical period, take the ratio of the difference to the average click rate in the statistical period as the increment of the click rate in the current sampling period, and determine the score value corresponding to the media information in the conversion evaluation dimension based on the increment, wherein when the increment is greater than 0, the score value corresponding to the media information in the conversion evaluation dimension is proportional to the increment, and when the increment is less than 0, the score value corresponding to the media information in the conversion evaluation dimension is zero; A risk determination module, configured to construct input features corresponding to the dimensions according to the evaluation results corresponding to the dimensions; use the input features and the image data of the media information as inputs of a combined model of a trained linear model and a neural network, respectively; and determine the risk type of the media information according to the output result of the combined model after inputting the first output of the linear model and the second output of the neural network into a logistic regression layer, wherein the evaluation result is characterized as a score value corresponding to the evaluation dimension; The recommendation review module is used to trigger the information delivery system to stop delivering the media information when the media information is of a set risk type.
13. A computer device, characterized in that: include: a processor and a memory for storing a computer program capable of being executed on the processor; Wherein, when the processor is used to run the computer program, it implements the information inspection method described in any one of claims 1 to 11.
14. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the information inspection method according to any one of claims 1 to 11 is implemented.
