Media data processing method and system, electronic equipment and storage medium

Abstract

The invention discloses a media data processing method and system, electronic equipment and a storage medium, and relates to the technical field of data processing. The method comprises the following steps: acquiring to-be-processed media data in an information streaming media scene; the to-be-processed media data are input into a value prediction model, a value distribution result is obtained through prediction, and the value prediction model is obtained through training of the media data samples and distribution results of value attribute samples of the media data samples; based on the value distribution result, a value attribute threshold value of the media file is determined, and the value attribute threshold value is used for enabling the income information of the media delivery platform to be larger than an income information threshold value; and based on the value attribute threshold, determining a target value attribute of the media file in the information streaming media scene, the target value attribute being greater than or equal to the value attribute threshold. The technical problem of low flexibility of media data processing is solved.

Description

Technical Field

[0001] The present application relates to the technical field of data processing, and in particular, to a method, a system, an electronic device, and a storage medium for processing media data. Background Art

[0002] Currently, when processing media data in the information flow media scenario, on the one hand, the assumption conditions of the existing auction theory do not match the actual application scenario, lacking practical implementation solutions; on the other hand, the fixed reserve price calculation method based on statistics cannot adapt to the dynamic characteristics of information flow advertisements, restricting the flexibility and efficiency of the media placement platform in media data processing and auction mechanism optimization, thus resulting in the problem of low flexibility in media data processing.

[0003] In view of the above problems, no effective solution has been proposed yet. Summary of the Invention

[0004] Embodiments of the present application provide a method, a system, an electronic device, and a storage medium for processing media data, so as to at least solve the technical problem of low flexibility in media data processing.

[0005] According to one aspect of the embodiments of the present application, a method for processing media data is provided. The method may include: obtaining media data to be processed in the information flow media scenario, where the media data to be processed at least includes a media file to be placed on a media placement platform; inputting the media data to be processed into a value prediction model to predict a value distribution result, where the value prediction model is trained using media data samples and the distribution results of value attribute samples of the media data samples, and the value distribution result is used to represent the distribution results of different value attributes of the media file; determining a value attribute threshold of the media file based on the value distribution result, where the value attribute threshold is used to make the revenue information of the media placement platform greater than the revenue information threshold; and determining the target value attribute of the media file in the information flow media scenario based on the value attribute threshold, where the target value attribute is greater than or equal to the value attribute threshold.

[0006] According to another aspect of the embodiments of the present application, another method for processing media data is provided. The method may include: obtaining the media data to be processed in the information stream media scenario from an e-commerce platform, where the media data to be processed at least includes a media file to be delivered to a media delivery platform; inputting the media data to be processed into a value prediction model to obtain a value distribution result, where the value prediction model is trained using media data samples and the distribution results of value attribute samples of the media data samples, and the value distribution result is used to represent the distribution results of different value attributes of the media file; determining a value attribute threshold of the media file based on the value distribution result, where the value attribute threshold is used to make the revenue information of the media delivery platform greater than the revenue information threshold; determining the target value attribute of the media file in the information stream media scenario based on the value attribute threshold, where the target value attribute is greater than or equal to the value attribute threshold; and sending the target value attribute to the e-commerce platform.

[0007] According to another aspect of the embodiments of the present application, yet another method for processing media data is provided. The method may include: obtaining the media data to be processed in the information stream media scenario by calling a first interface, where the first interface includes a first parameter, and the parameter value of the first parameter includes the media data to be processed, and the media data to be processed at least includes a media file to be delivered to a media delivery platform; inputting the media data to be processed into a value prediction model to obtain a value distribution result, where the value prediction model is trained using media data samples and the distribution results of value attribute samples of the media data samples, and the value distribution result is used to represent the distribution results of different value attributes of the media file; determining a value attribute threshold of the media file based on the value distribution result, where the value attribute threshold is used to make the revenue information of the media delivery platform greater than the revenue information threshold; determining the target value attribute of the media file in the information stream media scenario based on the value attribute threshold, where the target value attribute is greater than or equal to the value attribute threshold; and outputting the target value attribute by calling a second interface, where the second interface includes a second parameter, and the parameter value of the second parameter includes the target value attribute.

[0008] According to another aspect of the embodiments of the present application, a media data processing system is provided. The system may include: a client for transmitting media data to be processed in an information stream media scenario, where the media data to be processed at least includes media files to be delivered to a media delivery platform; a server for inputting the media data to be processed into a value prediction model to obtain a value distribution result, where the value prediction model is trained using media data samples and the distribution results of value attribute samples of the media data samples, and the value distribution result is used to represent the distribution results of different value attributes of the media files; determining a value attribute threshold for the media files based on the value distribution result, where the value attribute threshold is used to make the revenue information of the media delivery platform greater than the revenue information threshold; determining the target value attribute of the media files in the information stream media scenario based on the value attribute threshold, where the target value attribute is greater than or equal to the value attribute threshold; and outputting the target value attribute to the client.

[0009] According to another aspect of the embodiments of the present application, a computing device is further provided. The computing device may include a memory and a processor: the memory is used to store an executable program, and the processor is used to run the program, and when the above program runs, it executes the media data processing method of any one of the above.

[0010] According to another aspect of the embodiments of the present application, an electronic device is further provided. The electronic device may include a memory and a processor: the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, and when the above computer-executable instructions are executed by the processor, the media data processing method of any one of the above is implemented.

[0011] According to another aspect of the embodiments of the present application, a computer-readable storage medium is further provided. The computer-readable storage medium includes a stored program, where when the program runs, it controls the device where the storage medium is located to execute the media data processing method of any one of the above.

[0012] According to another aspect of the embodiments of the present application, a computer program product is further provided, including a non-volatile computer-readable storage medium. The non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for processing media data of any one of the above is implemented.

[0013] In an embodiment of the present application, media data to be processed in an information flow media scenario is obtained, where the media data to be processed at least includes media files to be delivered to a media delivery platform; the media data to be processed is input into a value prediction model, and a value distribution result is predicted. The value prediction model is trained using media data samples and the distribution results of value attribute samples of the media data samples. The value distribution result is used to represent the distribution results of different value attributes of the media files; based on the value distribution result, a value attribute threshold of the media files is determined, where the value attribute threshold is used to make the revenue information of the media delivery platform greater than the revenue information threshold; based on the value attribute threshold, the target value attribute of the media files in the information flow media scenario is determined, where the target value attribute is greater than or equal to the value attribute threshold. That is to say, in the information flow media scenario, the embodiment of the present application uses the value prediction model to predict the media data to be processed, obtains the value distribution result, and then uses the value distribution result to calculate in real time the value attribute threshold for the media data. And this value attribute threshold is a personalized dynamic reserve price, rather than the traditional fixed reserve price. Finally, at the billing stage, it is ensured that the target value attribute is not lower than the value attribute threshold, so as to achieve the goal of maximizing the revenue of the media delivery platform, and thus achieve the technical effect of improving the flexibility of media data processing, and solve the technical problem of low flexibility in media data processing.

[0014] It is easy to notice that the above general description and the following detailed description are only for exemplifying and explaining the present application, and do not constitute a limitation to the present application. Brief Description of the Drawings

[0015] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The schematic embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation to the present application. In the drawings:

[0016] Figure 1 is a schematic diagram of an application scenario of a method for processing media data according to an embodiment of the present application;

[0017] Figure 2 is a flowchart of a method for processing media data according to an embodiment of the present application;

[0018] Figure 3 is a flowchart of another method for processing media data according to an embodiment of the present application;

[0019] Figure 4 is a flowchart of yet another method for processing media data according to an embodiment of the present application;

[0020] Figure 5 is a schematic diagram of a system for processing media data according to an embodiment of the present application;

[0021] Figure 6 is a flowchart of an optional information flow advertisement auction according to an embodiment of the present application;

[0022] Figure 7 is a flowchart of an information flow advertisement auction with a dynamic reserve price according to an embodiment of the present application;

[0023] Figure 8 is a schematic structural diagram of an advertiser value distribution estimation model according to an embodiment of the present application;

[0024] Figure 9 is a structural block diagram of a computing environment of a method for processing media data according to an embodiment of the present application;

[0025] Figure 10 is a schematic diagram of a device for processing media data according to an embodiment of the present application;

[0026] Figure 11 is a schematic diagram of another device for processing media data according to an embodiment of the present application;

[0027] Figure 12 is a schematic diagram of yet another device for processing media data according to an embodiment of the present application;

[0028] Figure 13 is a structural block diagram of a computing device according to an embodiment of the present application;

[0029] Figure 14 is a structural block diagram of an electronic device according to an embodiment of the present application. Detailed implementation manners

[0030] In order to enable those skilled in the art to better understand the solutions of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0031] It should be noted that the terms "first", "second", etc. in the description, claims and the above-mentioned drawings of this application are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of this application described here can be implemented in an order other than those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0032] First, some nouns or terms that appear in the process of describing the embodiments of this application are applicable to the following explanations:

[0033] The advertising auction mechanism is a rule for the allocation and billing of online advertisements, which can be used to stipulate how advertising slots are allocated to advertisers participating in the auction, and how much advertisers need to pay to obtain these advertising slots. The advertising auction mechanism should consider both the revenue goals of the advertising platform and balance the user experience and the interests of advertisers;

[0034] The revenue maximization mechanism is an online advertising allocation and billing rule with the goal of maximizing the platform's revenue. By adjusting the auction rules, such as reserve prices, bidding strategies, advertising slot allocation logics, etc., to ensure that the advertising platform obtains the maximum possible economic benefits from online advertisements;

[0035] The advertiser value distribution prediction model is a model for personalized prediction of the advertiser value distribution at the traffic granularity, which can be used to estimate the true value of advertisers for advertising slots or user groups based on historical data and current market conditions;

[0036] The reserve price refers to the billing value not lower than the reserve price, which defines the billing price of the advertising platform for the advertising slot. In an auction, when an advertiser's bid is equal to or higher than the reserve price, the advertiser may obtain the advertising slot;

[0037] The dynamic reserve price means that the reserve price changes with the change of the traffic value, rather than a fixed value, and the value of the reserve price can be adjusted according to real-time market dynamics and the specific value of the advertising slot. In information flow advertising, the dynamic reserve price can be used to perform personalized evaluation on each piece of traffic or each advertising display opportunity based on the advertiser value distribution prediction model, so as to calculate the most suitable reserve price for the current situation.

[0038] The above-mentioned method for processing media data provided by the embodiments of this application can be applied to the Figure 1 application scenarios shown, but not limited thereto.Figure 1 It is a schematic diagram of an application scenario of a method for processing media data according to an embodiment of the present application. In the application scenario as shown in Figure 1 the figure, the server 10 can be a cloud. The server 10 can be connected to one or more client devices 20 through a local area network connection, a wide area network connection, an Internet connection, or other types of data networks. Here, the client devices 20 can include, but are not limited to: smart phones, tablet computers, laptop computers, palm computers, personal computers, smart home devices, vehicle-mounted devices, etc. The client devices together constitute a client opposite to the server. An operation interface for obtaining media data to be processed can be deployed on the graphical user interface of the client device. The client device 20 can interact with the user through the graphical user interface to implement the method for processing media data provided by the embodiment of the present application.

[0039] In the embodiment of the present application, the system composed of the client device 20 and the server 10 can perform the following steps: Perform corresponding operations in the operation interface on the client device 20 to obtain media data to be processed. The client device can obtain the media data to be processed and send it to the server through the network. After receiving the media data to be processed, the server can perform the following steps: Step S102, obtain the media data to be processed in the information flow media scenario, where the media data to be processed at least includes media files to be delivered to the media delivery platform; Step S104, input the media data to be processed into the value prediction model, and predict a value distribution result, where the value prediction model is trained using media data samples and the distribution results of value attribute samples of the media data samples, and the value distribution result is used to represent the distribution results of different value attributes of the media files; Step S106, based on the value distribution result, determine the value attribute threshold of the media file, where the value attribute threshold is used to make the revenue information of the media delivery platform greater than the revenue information threshold; Step S108, based on the value attribute threshold, determine the target value attribute of the media file in the information flow media scenario, where the target value attribute is greater than or equal to the value attribute threshold.

[0040] Under the above operating environment, the present application provides a method for processing media data as shown in Figure 2 the figure. Figure 2 It is a flowchart of a method for processing media data according to an embodiment of the present application. As shown in Figure 2 the figure, the method can include the following steps:

[0041] Step S202, obtain the media data to be processed in the information flow media scenario.

[0042] In the technical solution provided in step S202 of the present application above, the media data to be processed at least includes media files to be delivered to the media delivery platform.

[0043] In this embodiment, media data to be processed in an information flow media scenario can be obtained. Among them, the information flow media scenario can be an information interaction scenario in a digital media environment where media information (types including but not limited to text, pictures, audio, video, etc.) is transmitted to the corresponding terminal device of the user in a continuous, dynamic, and personalized manner. In this information interaction scenario, the corresponding terminal device of the user can perform relevant operation behaviors on the above media information. Optionally, in the above information flow media scenario, when the corresponding terminal device of the user browses or receives continuous media files, the media delivery platform dynamically adjusts the display and sorting of the media files based on user preferences, media value prediction, etc., to optimize user engagement and platform revenue. Optionally, the media data to be processed in the information flow media scenario can be continuously presented to the user in chronological order, user behavior, or algorithmic decision-making. In the information flow media scenario, the media files can be dynamically updated, and the user can interact with the media files by scrolling, clicking, swiping, etc.

[0044] For example, the information flow media scenario can be an information flow advertising scenario. For instance, it can be an advertising auction scenario on a social media platform. Since the advertising spaces on the social media platform are limited, advertisers compete for display opportunities through an auction mechanism, and the social media platform dynamically adjusts the advertisement display based on advertisement value and user preferences; the information flow media scenario can also be an information flow e-commerce scenario. For example, it can be a commodity recommendation auction scenario on an e-commerce platform. The e-commerce platform determines the display order in the commodity information flow through an auction mechanism, and advertisers can obtain a more prominent display position by bidding; the information flow media scenario can also be the information flow scenario of a news aggregation platform. When a user browses news on the news platform, the news platform can, based on factors such as user preferences, news popularity, and advertisers' bids, adjust the display order of news and advertisements in real time to maximize the user reading experience and the platform's advertising revenue; the information flow media scenario can also be the information flow scenario of a video platform. The video content on the video platform can be personalized sorted according to user preferences and viewing history, and video advertisements of advertisers will be inserted into the video stream. The video platform dynamically adjusts the display timing and frequency of the advertisements by analyzing user behavior and advertisement value to optimize the user viewing experience and advertising revenue. It should be noted that the above different information flow media scenarios are only for illustrative purposes and do not specifically limit the types of information flow media scenarios.

[0045] In this embodiment, the media data to be processed may at least include media files to be delivered by a media delivery object to a media delivery platform. The media delivery object may be an entity that expects to display media files on the media delivery platform. For example, the media delivery object may be an advertiser. The media file may be media material provided by the media delivery object for display. For example, the media file may be an advertisement. The media delivery platform may be a platform that provides media file display services. For example, the media delivery platform may be an advertising platform. It should be noted that the above are only examples and do not specifically limit the types of media delivery objects, media files, and media delivery platforms.

[0046] For example, the media data to be processed in the information stream media scenario can be obtained. The media data to be processed can be a <session, ad> pair (<session, ad> pair), which can represent the display opportunity of an advertisement in a specific user session. The <session, ad> pair can include information about the user and the current behavior (from the session), as well as specific information about the advertisement to be displayed (from the advertisement). By analyzing the <session, ad> pair, the advertising system can more accurately evaluate the potential value of this display opportunity for the advertiser.

[0047] Step S204: Input the media data to be processed into the value prediction model to obtain a value distribution result.

[0048] In the technical solution provided in step S204 of the present application, the value prediction model is trained using media data samples and the distribution results of value attribute samples of the media data samples. The value distribution result is used to represent the distribution results of different value attributes of the media file.

[0049] In this embodiment, after obtaining the media data to be processed in the information stream media scenario, the obtained media data to be processed can be input into the value prediction model to obtain a value distribution result. Among them, the value prediction model can be trained using media data samples and the distribution results of value attribute samples of the media data samples. For example, the value prediction model can be an advertiser value distribution estimation model, and the distribution results of value attribute samples of the media data samples can be used to represent the distribution of advertiser value, which can be represented by v.

[0050] In this embodiment, the value distribution result can be the probability distribution of different values that the media delivery object is willing to pay for the display opportunity of the media file, and can be used to represent the distribution results of different value attributes of the media file. For example, the value distribution result can be the result of the advertiser value distribution estimation model estimating the media data to be processed, and can also be called the advertiser value distribution probability density function, which can be represented by f(v). The value attribute can be used to characterize different prices of the media file.

[0051] For example, after obtaining the media data <session, ad> pair in the information flow media scenario, the <session, ad> pair can be input into the advertiser value distribution prediction model, and the advertiser value distribution probability density function f(v) of each <session, ad> pair can be predicted by using the advertiser value distribution prediction model, so as to obtain the value distribution result.

[0052] Step S206: Determine the value attribute threshold of the media file based on the value distribution result.

[0053] In the technical solution provided in step S206 of the present application above, the value attribute threshold is used to make the revenue information of the media placement platform greater than the revenue information threshold.

[0054] In this embodiment, after inputting the media data to be processed into the value prediction model and predicting the value distribution result, based on the predicted value distribution result, the value attribute threshold of the media file can be determined. Among them, the value attribute threshold can be used to make the revenue information of the media placement platform greater than the revenue information threshold, and can also be called the personalized reserve price, which can be represented by r*. The revenue information threshold can be a critical value set in advance according to the actual situation for measuring the revenue information.

[0055] Optionally, after inputting the media data <session, ad> pair into the advertiser value distribution prediction model and obtaining the value distribution result f(v), the personalized reserve price r* can be calculated according to the value distribution result f(v), and the calculated personalized reserve price r* can be used to make the revenue information of the advertising platform greater than the revenue information threshold, so as to meet the revenue maximization mechanism.

[0056] Step S208: Determine the target value attribute of the media file in the information flow media scenario based on the value attribute threshold.

[0057] In the technical solution provided in step S208 of the present application above, the target value attribute is greater than or equal to the value attribute threshold.

[0058] In this embodiment, after determining the value attribute threshold of the media file based on the value distribution result, based on the determined value attribute threshold, the target value attribute of the media file in the information flow media scenario can be determined. Among them, the target value attribute is greater than or equal to the value attribute threshold, and the target value attribute can be the billing price in the billing stage.

[0059] Optionally, after determining the value attribute threshold of the media file based on the value distribution result, based on the determined value attribute threshold, media files with bids lower than the value attribute threshold can be filtered. Further based on the value attribute threshold, the target value attribute of the filtered media file in the information stream media scenario can be determined.

[0060] Through the above steps S202 to S208 of this application, the media data to be processed in the information stream media scenario is obtained, where the media data to be processed at least includes the media files to be delivered to the media delivery platform; the media data to be processed is input into the value prediction model, and the value distribution result is predicted, where the value prediction model is trained using the media data sample and the distribution result of the value attribute sample of the media data sample, and the value distribution result is used to represent the distribution result of different value attributes of the media file; based on the value distribution result, the value attribute threshold of the media file is determined, where the value attribute threshold is used to make the revenue information of the media delivery platform greater than the revenue information threshold; based on the value attribute threshold, the target value attribute of the media file in the information stream media scenario is determined, where the target value attribute is greater than or equal to the value attribute threshold. That is to say, in the information stream media scenario, this application embodiment uses the value prediction model to predict the media data to be processed, obtains the value distribution result, and then uses the value distribution result to calculate the value attribute threshold for the media data in real time. And this value attribute threshold is a personalized dynamic reserve price, rather than the traditional fixed reserve price. Finally, in the billing stage, it is ensured that the target value attribute is not lower than the value attribute threshold, so as to achieve the goal of maximizing the revenue of the media delivery platform, and further achieve the technical effect of improving the flexibility of media data processing, and solve the technical problem of low flexibility of media data processing.

[0061] The above value distribution result of this embodiment will be further introduced below.

[0062] As an optional implementation manner, the value distribution result includes the distribution result of the transaction probability of the media file under different value attributes. The transaction probability is used to represent the degree of possibility that the media delivery object performs a transaction operation on the media file according to the corresponding value attribute. The operation result corresponding to the transaction operation is used to allow the media delivery object to deliver the media file on the media delivery platform.

[0063] In this embodiment, when the media data to be processed is input into the value prediction model, the predicted value distribution result may include the distribution result of the transaction probability of the media file under different value attributes. The transaction probability can be used to represent the degree of possibility that the media delivery object performs a transaction operation on the media file according to the corresponding value attribute. For example, the transaction probability can be used to represent the probability that the advertiser actually conducts a transaction (such as paying advertising fees) under the condition of a given value attribute (such as click-through rate).

[0064] Optionally, the operation result corresponding to the transaction operation can be used to allow the media placement object to place media files on the media placement platform. For example, based on the distribution result of the transaction probability, the advertising platform can set corresponding screening and billing strategies. For example, the advertising platform can set a reserve price. When the advertiser's bid is higher than or equal to the reserve price, and according to the analysis of the predicted transaction probability distribution, the possibility that the advertiser pays according to the corresponding value attribute is high enough, the advertisement is allowed to be placed, so as to ensure the maximization of the platform's revenue, while reducing the display of low-quality advertisements and improving the user experience.

[0065] The method for determining the value attribute threshold of the media file based on the value distribution result of this embodiment will be further introduced below.

[0066] As an optional implementation manner, step S206, determining the value attribute threshold of the media file based on the value distribution result, includes: converting the value distribution result into a virtual valuation function, where the virtual valuation function satisfies the monotonic change rule; determining the value attribute threshold based on the virtual valuation function.

[0067] In this embodiment, after inputting the media data to be processed into the value prediction model and predicting the value distribution result, the predicted value distribution result can be converted into a virtual valuation function. Further, based on the converted virtual valuation function, the value attribute threshold can be determined. Among them, the virtual valuation function can be represented by φ(υ), and the virtual valuation function satisfies the monotonic change rule. For example, the virtual valuation function satisfies the property of monotonic increase.

[0068] Optionally, the above value attribute threshold can be zero. Optionally, after converting the predicted value distribution result into a virtual valuation function, since the virtual valuation function φ(υ) corresponding to the lognormal distribution function satisfies the property of monotonic increase, the zero point r of φ(υ) can be obtained by the bisection method. * =φ -1 (0).

[0069] This embodiment can design and optimize the auction rules based on the estimated result of the advertiser's value distribution by calculating the virtual valuation function to achieve the maximization of the advertising platform's revenue. The virtual valuation function can be used to ensure the incentive compatibility (IC) of the auction mechanism, that is, to encourage the advertiser to report its true value, thereby improving the efficiency and fairness of the auction.

[0070] The method for converting the value distribution result into a virtual valuation function in this embodiment will be further introduced below.

[0071] As an alternative implementation, the value distribution result is described by a probability density function in a normal distribution state. Converting the value distribution result into a virtual valuation function includes: determining the cumulative distribution function corresponding to the probability density function, where the cumulative distribution function is used to represent the cumulative result of the distribution results of different value attributes and is in a normal distribution state; converting the cumulative distribution function and the probability density function into a virtual valuation function.

[0072] In this embodiment, after inputting the media data to be processed into the value prediction model and predicting the value distribution result, since the predicted value distribution result can be described by a probability density function in a normal distribution state, the cumulative distribution function corresponding to the probability density function can be determined. After determining the cumulative distribution function corresponding to the probability density function, the determined cumulative distribution function and the probability density function can be converted into a virtual valuation function. Among them, the cumulative distribution function can be used to represent the cumulative result of the distribution results of different value attributes and is in a normal distribution state. The cumulative distribution function can be represented by F(υ).

[0073] Optionally, after predicting the value distribution result, since the value distribution result can be described by a probability density function in a normal distribution state, the cumulative distribution function corresponding to the probability density function can be determined. Furthermore, the cumulative distribution function and the probability density function can be converted into a virtual valuation function through the following formula:

[0074]

[0075] Among them, φ(υ) can be used to represent the virtual valuation function, f(υ) can be used to represent the log-normal distribution probability density function estimated from the advertiser value distribution, F(υ) can be used to represent the cumulative distribution function of the log-normal distribution, and v can be used to represent the distribution of the advertiser value.

[0076] This embodiment converts the cumulative distribution function and the probability density function into a virtual valuation function, which is an important step in optimizing the information flow advertising auction algorithm. It can enable the advertising platform to achieve a more efficient, fair, and personalized auction mechanism, thereby improving revenue while optimizing the user experience and advertising quality.

[0077] Next, the method for determining the value attribute threshold based on the virtual valuation function in this embodiment will be further introduced.

[0078] As an alternative implementation, determining the value attribute threshold based on the virtual valuation function includes: determining the input value that makes the output result of the virtual valuation function zero; determining the input value as the value attribute threshold.

[0079] In this embodiment, after converting the value distribution result into a virtual valuation function, an input value for making the output result of the virtual valuation function zero can be determined. Further, this input value can be determined as the value attribute threshold. Wherein, the input value can be the zero point of the virtual valuation function.

[0080] Optionally, after converting the value distribution result into a virtual valuation function φ(υ), the input value for making the output result of the virtual valuation function φ(υ) zero can be determined by the bisection method, that is, the zero point of the virtual valuation function φ(υ) is determined, and then this zero point is determined as the value attribute threshold r*.

[0081] In this embodiment, the zero point of the virtual valuation function is obtained by the bisection method and used as the value attribute threshold (personalized reserve price), which can not only significantly improve the revenue of the advertising platform, but also ensure the fairness of the auction mechanism, market efficiency and optimization of user experience.

[0082] Next, a further introduction is made to the method of inputting the media data to be processed into the value prediction model in this embodiment to predict the value distribution result.

[0083] As an alternative implementation manner, step S204, inputting the media data to be processed into the value prediction model to predict the value distribution result, includes: extracting feature data from the media data to be processed, where the feature data is used to represent the attributes of the media file and / or the attributes of the media placement object, and the media placement object is used to place the media file on the media placement platform; inputting the feature data into the value prediction model, and using the mean and variance in the value prediction model to convert the feature data into a value distribution result that satisfies the normal distribution state, where the mean and variance match the normal distribution state.

[0084] In this embodiment, after obtaining the media data to be processed in the information flow media scenario, feature data can be extracted from the obtained media data to be processed. Further, the extracted feature data can be input into a value prediction model, and by using the mean and variance in the value prediction model, the feature data is converted into a value distribution result that satisfies the normal distribution state. Among them, the mean and variance match the normal distribution state. The feature data can be used to represent the attributes of the media file and / or the attributes of the media placement object. For example, the feature data can include the attribute features on the advertising side, the predicted value (Predicted Value, abbreviated as pValue) features on the advertising side, the statistical features on the advertising side, and the page identifier (PageIdentifier, abbreviated as pid) features on the request side, etc. The attribute features on the advertising side can be used to describe various attributes of the advertisement itself, such as the type, size, format, etc. of the advertisement. The pValue features on the advertising side can be used to represent the predicted values of various user behaviors. For example, the pValue features can be the predicted click-through rate (Click-Through Rate, abbreviated as CTR), conversion rate (Conversion Rate, abbreviated as CVR), purchase rate (Purchase Rate, abbreviated as PUR), etc. The statistical features on the advertising side can include the statistical information of the advertising historical performance, such as the number of clicks, number of displays, number of conversions, average click-through rate, average conversion rate, etc. of the advertisement in the past period of time. The pid features on the request side can be used to represent the source page information of the advertisement request, such as the page type, page theme, location of the page in the website or application, etc.

[0085] Optionally, after obtaining the media data to be processed in the information flow media scenario, feature data such as the attribute features on the advertising side, the pValue features on the advertising side, the statistical features on the advertising side, and the pid features on the request side can be extracted from the obtained media data to be processed. It should be noted that the information on the user side is already included in the pValue features on the advertising side, and there is no need to repeat the modeling. After extracting the feature data from the media data to be processed, the extracted feature data can be input into a value prediction model, and by using the mean and variance in the value prediction model, the feature data is converted into a value distribution result that satisfies the normal distribution state.

[0086] By selecting feature data such as the attribute features on the advertising side, the pValue features on the advertising side, the statistical features on the advertising side, and the pid features on the request side in this embodiment, and then converting the feature data into a value distribution result, the advertising platform can make decisions based on data rather than relying on intuition or experience, thereby providing a more scientific and rigorous basis for the auction mechanism design of the advertising platform.

[0087] As an alternative implementation, the media data sample to be processed at least includes a media file sample to be delivered to a media delivery platform, and the method further includes: extracting a feature data sample from the media data sample, where the feature data sample is used to represent the attributes of the media file sample and / or the attributes of the media delivery object sample, and the media delivery object sample is used to deliver the media file sample to the media delivery platform; respectively determining distribution result samples of different value attribute samples of the media file sample under different dimensions of feature data in the feature data sample; in response to the distribution result sample being in a normal distribution state, determining a mean and a variance that match the normal distribution state; and training a value prediction model using the mean and the variance.

[0088] In this embodiment, the media data sample to be processed may at least include a media file sample to be delivered by a media delivery object sample to the media delivery platform. A feature data sample may be extracted from the media data sample, and further, distribution result samples of different value attribute samples of the media file sample may be respectively determined under different dimensions of feature data in the feature data sample. When the distribution result sample is in a normal distribution state, a mean and a variance that match the normal distribution state may be determined. After determining the mean and the variance that match the normal distribution state, a value prediction model may be trained using the determined mean and variance. Among them, the feature data sample may be used to represent the attributes of the media file sample and / or the attributes of the media delivery object sample. For example, the feature data sample may include an attribute feature sample on the advertising side, a pValue feature sample on the advertising side, a statistical feature sample on the advertising side, a pid feature sample on the request side, etc. The feature data sample may also be referred to as a feature vector and may be represented by x, and x may be represented as {x1, x2,..., xn}, and x1 may be used to represent a certain dimension feature of the feature vector x.

[0089] Optionally, after extracting the feature data sample from the media data sample, that is, after extracting the feature vector x, distribution result samples of different value attribute samples of the media file sample may be respectively determined under different dimensions of feature data in the feature data sample. That is, the distribution v of the advertiser value may be determined under the given feature vector x, that is, the conditional probability p(v|x) may be determined.

[0090] It should be noted that the distribution form of the conditional probability p(v|x) for a given feature vector x is unknown. A sample of traffic granularity is a sampling from this distribution, and its distribution cannot be observed from the posterior data. Therefore, in this embodiment, the frequency histogram of the posterior value v is first observed to approximately replace the probability distribution, that is, p(v), and it is observed that the probability distribution p(v) basically follows a lognormal distribution. Further, by dividing different dimensions and observing the frequency histogram of p(v), in fact, the conditional probability distributions of p(v|x1 = C1) and p(v|x1 = C2) are observed (where C1 and C2 can be used to represent different values of the feature x1), and it is observed that this conditional probability distribution also basically follows a lognormal distribution. Therefore, the distribution result sample (i.e., the distribution to be estimated) can be assumed to be a lognormal distribution, which can be expressed as:

[0091] υ~LogN(μ(x,θ),σ 2 (x,θ))

[0092] where can be used to represent the expectation of the distribution result sample, μ can be used to represent the mean of the estimated lognormal distribution, σ 2 can be used to represent the variance of the estimated lognormal distribution, x can be used to represent the input feature, θ can be used to represent the model parameter, and N can be used to represent the normal distribution. After determining the mean and variance that match the normal distribution state, the value prediction model can be trained using the determined mean and variance.

[0093] Through the observation of the frequency histogram of the posterior value, this embodiment finds that it basically follows a lognormal distribution. Therefore, the distribution of the advertiser value to be estimated is assumed to be a lognormal distribution, and then the mean and variance that match the normal distribution state are determined to train the value prediction model.

[0094] As an alternative implementation, the method further includes: determining different operation behavior data corresponding to the media file samples in the feature data samples, where the operation behavior data is used to represent the operation behavior on the media file samples; combining the different operation behavior data to obtain a combination result; in response to the distribution result sample being in a normal distribution state, determining the mean and variance that match the normal distribution state, including: in response to the distribution result sample being in a normal distribution state, using the combination result to determine the mean and variance that match the normal distribution state.

[0095] In this embodiment, different operation behavior data corresponding to media file samples in the feature data samples can be determined. Further, different operation behavior data can be combined to obtain a combined result. When the distribution result sample is in a normal distribution state, the obtained combined result can be used to determine the mean and variance that match the normal distribution state. Among them, the operation behavior data can be used to represent the operation behavior on the media file sample, and can also be called the pValue feature.

[0096] Optionally, after determining the different operation behavior data corresponding to the media file samples in the feature data samples, that is, after determining the pValue feature, the different pValue features can be combined by the following formula to obtain a combined result (that is, other pre-estimated values with physical meanings):

[0097] pctcvr = pctr * pcvr

[0098] Among them, pctcvr can be used to represent the probability of a deal after a certain product is exposed. pctr can be used to represent the click probability predicted by the refined ranking click-through rate prediction model. pcvr can be used to represent the probability of whether a deal will be made under the condition of a click predicted by the refined ranking conversion rate prediction model. Among them, the goal of the refined ranking click-through rate prediction model is to predict the probability that a specific advertisement will be clicked when shown to a user. This model usually learns a probability function based on historical data, including but not limited to user characteristics, advertisement characteristics, context characteristics (such as display page type, timestamp, etc.). This function inputs an advertisement and a user (and possibly context information) and outputs the probability that the advertisement will be clicked by the user. The refined ranking click-through rate prediction model is an important basis for calculating the advertisement value and determining the advertisement display order, and can help distinguish which advertisements are more likely to attract user interaction. The purpose of the refined ranking conversion rate prediction model is to predict the probability of finally completing a certain conversion behavior (such as purchase, registration, form submission, etc.) after the user clicks on the advertisement. In other words, the refined ranking conversion rate prediction model calculates the conversion probability after a click. This model can be based on advertisement historical data and user behavior data, input features including advertisement characteristics, user characteristics, and click behavior, and output the probability of conversion occurrence.

[0099] Optionally, for the understanding of a certain linear relationship with the pValue feature, the embedding representation of the identifier (Identifier, abbreviated as id) feature is multiplied element-wise with the pValue feature to determine the mean and variance that match the normal distribution state.

[0100] Optionally, after selecting the attribute features on the advertisement side, the pValue features on the advertisement side, the statistical features on the advertisement side, and the pid features on the request side, etc. as input features, the above feature data is combined and specially processed according to expert experience. For example, the embedding of the id feature is element-wise multiplied by the pValue, and finally the mean and variance that match the normal distribution state are determined.

[0101] In this embodiment, the embedding of the id feature is element-wise multiplied by the pValue, which can adjust the importance of each dimension in the embedding vector. If the pValue of a certain dimension is small, it means that the information on this dimension is relatively important for the current task. Therefore, the corresponding weight will increase accordingly, so that the statistically significant information can be directly incorporated into the model, which helps to improve the model's attention to certain key features and also avoids the risk of overfitting.

[0102] Next, the method of training the value prediction model using the mean and variance in this embodiment will be further introduced.

[0103] As an optional implementation, training a value prediction model using the mean and variance includes: using the mean and variance to establish a loss function; using the loss function to determine the model parameters of the value prediction model to be generated; establishing a value prediction model including the mean, variance, and model parameters.

[0104] In this embodiment, after determining the mean and variance that match the normal distribution state in response to the distribution result sample being in the normal distribution state, the determined mean and variance can be used to establish a loss function. After using the mean and variance to establish a loss function, the established loss function can be used to determine the model parameters θ of the value prediction model to be generated. Further establish a value prediction model including the mean μ, variance σ

[0105] Optionally, after determining the mean and variance that match the normal distribution state in response to the distribution result sample being in the normal distribution state, the loss function can be derived from the maximum likelihood estimation:

[0106]

[0107] where L Lognormal can be used to represent the loss function based on the lognormal distribution. Further, after using the mean and variance to establish a loss function, the established loss function can be used to determine the model parameters θ of the value prediction model to be generated. Further establish a value prediction model including the mean μ, variance σ 2 and model parameters θ.

[0108] After assuming that the advertiser value distribution to be estimated follows a lognormal distribution, this embodiment further derives a loss function from maximum likelihood estimation for training the advertiser value distribution estimation model.

[0109] The following further introduces the method of this embodiment for determining the target value attribute of a media file in the in-feed media scenario based on the value attribute threshold.

[0110] As an optional implementation manner, step S208 of determining the target value attribute of a media file in the in-feed media scenario based on the value attribute threshold includes: determining the first initial value attribute of the media file in the in-feed media scenario; determining the maximum value attribute between the first initial value attribute and the value attribute threshold; and determining the value attribute greater than or equal to the maximum value attribute as the target value attribute.

[0111] In this embodiment, after determining the value attribute threshold of the media file based on the value distribution result, the first initial value attribute of the media file in the in-feed media scenario can be determined. Further, the maximum value attribute can be determined between the first initial value attribute and the value attribute threshold. After determining the maximum value attribute, the value attribute greater than or equal to the maximum value attribute can be determined as the target value attribute. Among them, the first initial value attribute can be used to represent that the settlement price is the minimum bid to maintain the current slot, and can also be referred to as critical bid. The maximum value attribute can be represented by settle_price.

[0112] Optionally, after determining the first initial value attribute, that is, determining the original critical bid, the maximum value attribute between the original critical bid and the value attribute threshold (i.e., the personalized reserve price r*) can be determined: settle_price = max(original critical bid, r*), and then the value attribute greater than or equal to the maximum value attribute is determined as the target value attribute.

[0113] By determining that the target value attribute is greater than or equal to the maximum value attribute (i.e., the settlement price is equal to the maximum value between the original critical bid and the personalized reserve price r*), taking the maximum value can ensure that the billing price is not lower than the personalized reserve price r*. Even under the second-price billing rule, the advertising platform can ensure that the billing for its ad slots is not lower than the value it expects for the ad slots, avoiding low-price transactions, thereby protecting and enhancing the actual value of the ad slots and the total revenue of the advertising platform.

[0114] As an alternative implementation, the method further includes: obtaining a second initial value attribute of the media file feedback by the media placement platform; determining the target value attribute of the media file in the information flow media scenario based on the value attribute threshold, including: in the case that the information flow media scenario is a first type of information flow media scenario, in response to the second initial value attribute being greater than or equal to the value attribute threshold, determining the target value attribute of the media file in the information flow media scenario based on the value attribute threshold, where the first type of information flow media scenario is used to represent an information flow media scenario including media content of the media file and not including natural content.

[0115] In this embodiment, the second initial value attribute of the media file feedback by the media placement platform can be obtained. Further, in the case that the information flow media scenario is a first type of information flow media scenario, when the second initial value attribute is greater than or equal to the value attribute threshold, the target value attribute of the media file in the information flow media scenario can be determined based on the value attribute threshold. Among them, the second initial value attribute can be used to represent the price of the media file feedback by the media placement platform. For example, the second initial value attribute can be used to represent the expected price of a certain traffic for competing ads considered by the advertising platform.

[0116] In this embodiment, the first type of information flow media scenario can be used to represent an information flow media scenario including media content of the media file and not including natural content. The above-mentioned first type of information flow media scenario can also be called a no-natural-card scenario. For example, the no-natural-card scenario can be used to represent that the display of an advertisement depends on paid promotion, that is, the advertiser needs to pay a fee to ensure that its content can be displayed to the target audience at a specific time and place.

[0117] Optionally, obtaining the second initial value attribute of the media file feedback by the media placement platform, that is, obtaining the expected price of this traffic for competing ads considered by the advertising platform. In the case that the information flow media scenario is a no-natural-card scenario, when the second initial value attribute is less than the value attribute threshold, that is, the bid price is lower than the personalized reserve price, it is considered that the competing ad bid is too low and should lose the competing opportunity, and the competing bids lower than the personalized reserve price can be filtered out. When the second initial value attribute is greater than or equal to the value attribute threshold, the target value attribute of the media file in the information flow media scenario can be determined based on the value attribute threshold.

[0118] The personalized reserve price in this embodiment is calculated based on the advertiser value distribution prediction model, which can reflect the expected value of the ad space in a certain auction considered by the advertising platform. By filtering out the bids lower than the reserve price, the advertising platform can avoid low-price transactions, thereby increasing the average transaction price of the ad space and directly promoting the increase of the advertising platform's revenue.

[0119] As an alternative implementation, the media file is from a media file queue, and the method further includes: determining the number of media files to be filtered in the media file queue whose second initial value attribute is less than the value attribute threshold; determining the ratio between the number of media files to be filtered and the number of media files in the media file queue; and in the case where the ratio is less than or equal to the ratio threshold, filtering out the media files to be filtered from the media file queue in response to the second initial value attribute being less than the value attribute threshold.

[0120] In this embodiment, the media file can be from a media file queue. In the media file queue, the number of media files to be filtered whose second initial value attribute is less than the value attribute threshold can be determined. Further, the ratio between the number of media files to be filtered and the number of media files in the media file queue can be determined. After determining the ratio between the number of media files to be filtered and the number of media files in the media file queue, in the case where the ratio is less than or equal to the ratio threshold, when the second initial value attribute is less than the value attribute threshold, the media files to be filtered are filtered out from the media file queue. Among them, the number of media files to be filtered whose second initial value attribute is less than the value attribute threshold can be the number of advertisements filtered by the personalized reserve price in the media file queue, which can be represented by M. The number of media files in the media file queue can be the length of the media file queue, which can be represented by N. The ratio threshold can be a critical value preset according to the actual situation for measuring the ratio, which can be represented by filter_ratio.

[0121] It should be noted that when the bid price is lower than the personalized reserve price, it is considered that the competing ad bid price is too low and the competing opportunity should be lost, and the competing ads lower than the personalized reserve price can be filtered out. However, complete filtering requires the advertising platform to bear a greater risk of auction failure. Consider a situation: due to serious overestimation by the model prediction, the bid prices of competing ads on a certain request are all lower than the personalized reserve price. After filtering, no ad is returned. If there is no fallback strategy at the front end, it will cause the user to see a blank screen, and the risk is extremely high.

[0122] Optionally, to solve the above problems, for stability considerations, this embodiment adopts a relatively conservative filtering strategy by setting a ratio threshold (i.e., a filtering ratio parameter) filter_ratio. Let the number of media files in the media file queue (i.e., the queue length) be N, and the number of media files to be filtered whose second initial value attribute is less than the value attribute threshold (i.e., the number of advertisements filtered by the reserve price in the queue) be M. If the ratio M / N between the number of media files to be filtered and the number of media files in the media file queue is less than or equal to the ratio threshold, that is, M / N <= filter_ratio, then the media files to be filtered are filtered out from the media file queue, that is, these M advertisements are filtered.

[0123] In this embodiment, the risk is controlled by setting the filtering ratio parameter, which can control the number of advertisements to be filtered, ensuring that while removing low-value advertisements, there are still enough advertisements shown to users, and avoiding problems such as a decline in user experience or vacant advertisement slots caused by excessive filtering.

[0124] As an alternative implementation, the method further includes: when the ratio is greater than the ratio threshold, retaining the media file to be filtered in the media file queue; setting the value attribute of the media file to be filtered to the default value attribute.

[0125] In this embodiment, after determining the ratio between the number of media files to be filtered and the number of media files in the media file queue, when the determined ratio is greater than the ratio threshold, the media file to be filtered can be retained in the media file queue, and further, the value attribute of the media file to be filtered can be set to the default value attribute. Among them, the default value attribute can be the default value of the value attribute set in advance according to the actual situation. For example, the default value attribute can be 0. This is only for illustrative purposes and does not specifically limit the value of the default value attribute.

[0126] Optionally, if the ratio M / N between the number of media files to be filtered and the number of media files in the media file queue is greater than the ratio threshold, that is, M / N > filter_ratio, the reserve price filtering logic does not take effect, the competing advertisements are retained, and the reserve price is set to 0.

[0127] In this embodiment, when M / N > filter_ratio, the reserve price filtering logic does not take effect, the competing advertisements are retained, and the reserve price is set to 0, which can avoid a large number of advertisements being filtered due to an overly high reserve price estimated by the model, thereby reducing the risk of auction failure (i.e., the advertisement slot cannot be filled).

[0128] Next, the method for determining the target value attribute of a media file in the information stream media scenario based on the value attribute threshold in this embodiment is further introduced.

[0129] As an alternative implementation, step S208, determining the target value attribute of a media file in the information stream media scenario based on the value attribute threshold, includes: when the information stream media scenario is a second type of information stream media scenario, transmitting the value attribute threshold to the fusion billing service of the media delivery platform, where the second type of information stream media scenario is used to represent an information stream media scenario including media content of media files and including natural content; in the fusion billing service, determining the target value attribute of the media file in the information stream media scenario based on the value attribute threshold.

[0130] In this embodiment, after determining the value attribute threshold of the media file based on the value distribution result, in the case where the information flow media scenario is the second type of information flow media scenario, the determined value attribute threshold can be transparently transmitted to the fusion billing service of the media placement platform. Further, in the fusion billing service, based on the determined value attribute threshold, the target value attribute of the media file in the information flow media scenario can be determined. The second type of information flow media scenario can be used to represent an information flow media scenario that includes media content of the media file and also includes natural content. The second type of information flow media scenario can also be referred to as a natural card scenario. For example, the natural card scenario can be used to represent that an advertisement is seen by a user in a non-paid and natural way, that is, the advertisement usually obtains a display opportunity due to its own quality, relevance, and the role of the platform recommendation mechanism.

[0131] It should be noted that there is no fusion stage in the natural cardless scenario, and the allocation result and the billing result are directly determined in the refined competition stage. In the natural card scenario, the final allocation result and the billing result are determined through the fusion stage.

[0132] Optionally, in the natural card scenario, the value attribute threshold (i.e., the personalized reserve price) can be transparently transmitted to the fusion billing service of the media placement platform. Optionally, by adding a personalized reserve price field to the return field of the Ad Intelligence Module (AIM for short), it is ensured that the fusion billing service can obtain the personalized reserve price for billing, thereby realizing the transparent transmission of the reserve price. Among them, AIM can be the central module of the advertising engine and can be used to interact with the recommendation engine. AIM returns advertisements to the recommendation engine, and the recommendation engine then calls the fusion service (i.e., the mixed arrangement module). Since the advertisement billing in the natural card scenario is processed in the mixed arrangement module, AIM needs to transmit the personalized reserve price calculated in the refined arrangement stage to the fusion service.

[0133] Through the transparent transmission of the personalized reserve price in this embodiment, the billing service can accurately use the reserve price of each advertisement for billing, avoiding the problem of inaccurate billing caused by using a unified or static reserve price, thereby ensuring the fairness and reasonableness of billing and reducing the billing error.

[0134] Next, the method for obtaining the media data to be processed in the information flow media scenario of this embodiment will be further introduced.

[0135] As an alternative implementation, in step S202, obtaining the media data to be processed in the information flow media scenario includes: in response to the generation of traffic by the media placement platform, obtaining the media data to be processed in the information flow media scenario; in step S206, determining the value attribute threshold of the media file based on the value distribution result includes: determining the value attribute threshold of the traffic for the media file based on the value distribution result.

[0136] In this embodiment, in response to the generation of traffic by the media placement platform, the media data to be processed in the information flow media scenario can be obtained. After inputting the media data to be processed into the value prediction model and predicting the value distribution result, based on the predicted value distribution result, the value attribute threshold of the traffic for the media file can be determined.

[0137] Optionally, in response to the generation of traffic by the media placement platform, the media data to be processed in the information flow media scenario can be obtained. Further inputting the media data to be processed at the traffic granularity into the value prediction model, the value prediction model predicts the possible payment willingness distribution of the advertiser based on the media data to be processed at the traffic granularity, that is, the value distribution result of the ad slot. According to the predicted value distribution result, the personalized reserve price for the current ad slot and traffic is calculated.

[0138] The value prediction model in this embodiment can be a model for personalized estimation of the advertiser value distribution at the traffic granularity, which can ensure that each ad auction generated by the traffic is dynamically priced based on the specific value of the traffic. This can not only increase the revenue of the ad platform, but also optimize the allocation efficiency of the ad slots. At the same time, through the setting of the reserve price, it encourages advertisers to bid according to the true value of the ad, thereby improving the market ecology and enhancing the overall ad quality and user experience.

[0139] In the embodiments of the present application, to obtain the media data to be processed in the information flow media scenario, where the media data to be processed at least includes the media files to be delivered to the media delivery platform; input the media data to be processed into the value prediction model, and a value distribution result is predicted. The value prediction model is trained using media data samples and the distribution results of the value attribute samples of the media data samples. The value distribution result is used to represent the distribution results of different value attributes of the media files; based on the value distribution result, determine the value attribute threshold of the media file, where the value attribute threshold is used to make the revenue information of the media delivery platform greater than the revenue information threshold; based on the value attribute threshold, determine the target value attribute of the media file in the information flow media scenario, where the target value attribute is greater than or equal to the value attribute threshold. That is to say, in the embodiments of the present application, in the information flow media scenario, the value prediction model is used to predict the media data to be processed to obtain the value distribution result, and then the value attribute threshold for the media data is calculated in real time using the value distribution result. And this value attribute threshold is a personalized dynamic reserve price, rather than the traditional fixed reserve price. Finally, in the billing stage, it is ensured that the target value attribute is not lower than the value attribute threshold, so as to achieve the goal of maximizing the revenue of the media delivery platform, and thus achieve the technical effect of improving the flexibility of media data processing, and solve the technical problem of low flexibility in media data processing.

[0140] The embodiments of the present application also provide a method for processing media data, Figure 3 which is a flowchart of another method for processing media data according to the embodiments of the present application, as Figure 3 shown. The method may include the following steps:

[0141] Step S302, obtain the media data to be processed in the information flow media scenario from the e-commerce platform.

[0142] In the technical solution provided in step S302 of the present application, the media data to be processed in the information flow media scenario can be obtained from the e-commerce platform. Among them, the media data to be processed at least includes the media files to be delivered to the media delivery platform.

[0143] For example, from the e-commerce platform, the media data to be processed in the information flow media scenario can be obtained. The media data to be processed can be a <session, ad> pair, which can represent the display opportunity of an advertisement in a specific user session. The <session, ad> pair can include information about the user and their current behavior (from the session), and specific information about the advertisement to be displayed (from the advertisement). By analyzing the <session, ad> pair, the advertising system can more accurately evaluate the potential value of this display opportunity for the advertiser.

[0144] Step S304: Input the media data to be processed into the value prediction model to obtain the value distribution result.

[0145] In the technical solution provided in step S304 of the present application, the value prediction model is trained using media data samples and the distribution results of value attribute samples of the media data samples. The value distribution result is used to represent the distribution results of different value attributes of media files.

[0146] In this embodiment, after obtaining the media data to be processed in the information flow media scenario from the e-commerce platform, the obtained media data to be processed can be input into the value prediction model to obtain the value distribution result.

[0147] For example, after obtaining the media data to be processed <session, ad> pairs in the information flow media scenario, the <session, ad> pairs can be input into the advertiser value distribution estimation model, and the advertiser value distribution probability density function f(v) can be estimated for each <session, ad> pair using the advertiser value distribution estimation model, thereby obtaining the value distribution result.

[0148] Step S306: Determine the value attribute threshold of the media file based on the value distribution result.

[0149] In the technical solution provided in step S306 of the present application, the value attribute threshold is used to make the revenue information of the media placement platform greater than the revenue information threshold.

[0150] In this embodiment, after inputting the media data to be processed into the value prediction model to obtain the value distribution result, based on the predicted value distribution result, the value attribute threshold of the media file can be determined.

[0151] Optionally, after inputting the media data to be processed <session, ad> pairs into the advertiser value distribution estimation model to obtain the value distribution result f(v), the personalized reserve price r* can be calculated according to the value distribution result f(v), and the calculated personalized reserve price r* can be used to make the revenue information of the advertising platform greater than the revenue information threshold, thereby satisfying the revenue maximization mechanism.

[0152] Step S308: Determine the target value attribute of the media file in the information flow media scenario based on the value attribute threshold.

[0153] In the technical solution provided in step S308 of the present application, the target value attribute is greater than or equal to the value attribute threshold.

[0154] In this embodiment, after determining the value attribute threshold of the media file based on the value distribution result, the target value attribute of the media file in the in-feed media scenario can be determined based on the determined value attribute threshold.

[0155] Optionally, after determining the value attribute threshold of the media file based on the value distribution result, media files with bids lower than the value attribute threshold can be filtered based on the determined value attribute threshold. Further based on the value attribute threshold, the target value attribute of the filtered media file in the in-feed media scenario can be determined.

[0156] Step S310: Send the target value attribute to the e-commerce platform.

[0157] In the technical solution provided in step S310 of the present application, after determining the target value attribute of the media file in the in-feed media scenario based on the value attribute threshold, the determined target value attribute can be sent to the e-commerce platform.

[0158] Through steps S302 to S310 of the present application, the to-be-processed media data in the in-feed media scenario is obtained from the e-commerce platform, where the to-be-processed media data at least includes media files to be delivered to the media delivery platform; the to-be-processed media data is input into the value prediction model to obtain a value distribution result, where the value prediction model is trained using media data samples and the distribution results of the value attribute samples of the media data samples, and the value distribution result is used to represent the distribution results of different value attributes of the media file; based on the value distribution result, the value attribute threshold of the media file is determined, where the value attribute threshold is used to make the revenue information of the media delivery platform greater than the revenue information threshold; based on the value attribute threshold, the target value attribute of the media file in the in-feed media scenario is determined, where the target value attribute is greater than or equal to the value attribute threshold; the target value attribute is sent to the e-commerce platform. That is to say, in the in-feed media scenario of the embodiment of the present application, the to-be-processed media data obtained from the e-commerce platform is predicted using the value prediction model to obtain a value distribution result, and then the value attribute threshold for the media data is calculated in real time using the value distribution result. Moreover, the value attribute threshold is a personalized dynamic reserve price rather than a traditional fixed reserve price. Finally, it is ensured that the target value attribute is not lower than the value attribute threshold during the billing stage, thereby achieving the goal of maximizing the revenue of the media delivery platform, and further realizing the technical effect of improving the flexibility of media data processing and solving the technical problem of low flexibility in media data processing.

[0159] Next, the method for determining the value attribute threshold of the media file based on the value distribution result in this embodiment will be further introduced.

[0160] As an alternative implementation, based on the value distribution result, determining the value attribute threshold of the media file includes: converting the value distribution result into a virtual valuation function, where the virtual valuation function satisfies the monotonic change rule; determining the value attribute threshold based on the virtual valuation function; the method further includes: sending the value attribute threshold to the e-commerce platform.

[0161] In this embodiment, after inputting the media data to be processed into the value prediction model and predicting the value distribution result, the predicted value distribution result can be converted into a virtual valuation function. Further, based on the converted virtual valuation function, the value attribute threshold is determined. Finally, the determined value attribute threshold is sent to the e-commerce platform. Among them, the virtual valuation function satisfies the monotonic change rule. For example, the virtual valuation function satisfies the property of monotonic increase.

[0162] Optionally, after converting the predicted value distribution result into a virtual valuation function, since the virtual valuation function φ(υ) corresponding to the lognormal distribution function satisfies the property of monotonic increase, the zero point of φ(υ) can be obtained by the bisection method, and this zero point is the value attribute threshold: r * = φ -1 (0).

[0163] This embodiment can design and optimize the auction rules based on the estimated value distribution result of the advertiser by calculating the virtual valuation function, so as to maximize the revenue of the advertising platform. The virtual valuation function can be used to ensure the incentive compatibility IC of the auction mechanism, that is, to encourage advertisers to report their true values, thereby improving the efficiency and fairness of the auction.

[0164] The embodiment of the present application also provides a method for processing media data, Figure 4 which is a flowchart of another method for processing media data according to the embodiment of the present application, as Figure 4 shown, the method may include the following steps:

[0165] Step S402, obtaining the media data to be processed in the information flow media scenario by calling the first interface.

[0166] In the technical solution provided in step S402 of the present application above, the media data to be processed in the information flow media scenario can be obtained by calling the first interface. Among them, the first interface may include a first parameter, and the parameter value of the first parameter may include the media data to be processed, and the media data to be processed may at least include the media file to be placed on the media placement platform.

[0167] For example, by invoking the first interface, the media data to be processed in the information flow media scenario can be obtained. The media data to be processed can be a <session, ad> pair, which can represent the display opportunity of an advertisement in a specific user session. The <session, ad> pair can include information about the user and their current behavior (from the session), as well as specific information about the advertisement to be displayed (from the advertisement). By analyzing the <session, ad> pair, the advertising system can more accurately evaluate the potential value of this display opportunity for the advertiser.

[0168] Step S404: Input the media data to be processed into the value prediction model to obtain a value distribution result.

[0169] In the technical solution provided in step S404 of the present application above, the value prediction model is trained using media data samples and the distribution results of value attribute samples of the media data samples. The value distribution result is used to represent the distribution results of different value attributes of the media file.

[0170] In this embodiment, after obtaining the media data to be processed in the information flow media scenario by invoking the first interface, the obtained media data to be processed can be input into the value prediction model to obtain a value distribution result.

[0171] For example, after obtaining the <session, ad> pair of the media data to be processed in the information flow media scenario, the <session, ad> pair can be input into the advertiser value distribution estimation model, and the advertiser value distribution probability density function f(v) can be estimated for each <session, ad> pair using the advertiser value distribution estimation model, thereby obtaining a value distribution result.

[0172] Step S406: Based on the value distribution result, determine the value attribute threshold of the media file.

[0173] In the technical solution provided in step S406 of the present application above, the value attribute threshold is used to make the revenue information of the media placement platform greater than the revenue information threshold.

[0174] In this embodiment, after inputting the media data to be processed into the value prediction model to obtain a value distribution result, based on the obtained value distribution result, the value attribute threshold of the media file can be determined.

[0175] Optionally, after inputting the <session, ad> pair of the media data to be processed into the advertiser value distribution estimation model to obtain a value distribution result f(v), the personalized reserve price r* can be calculated according to the value distribution result f(v), and the calculated personalized reserve price r* can be used to make the revenue information of the advertising platform greater than the revenue information threshold, thereby satisfying the revenue maximization mechanism.

[0176] Step S408: Determine the target value attribute of the media file in the information flow media scenario based on the value attribute threshold.

[0177] In the technical solution provided in step S408 of the present application, the target value attribute is greater than or equal to the value attribute threshold.

[0178] In this embodiment, after determining the value attribute threshold of the media file based on the value distribution result, the target value attribute of the media file in the information flow media scenario can be determined based on the determined value attribute threshold.

[0179] Optionally, after determining the value attribute threshold of the media file based on the value distribution result, the media files with bids lower than the value attribute threshold can be filtered based on the determined value attribute threshold. Further, based on the value attribute threshold, the target value attribute of the filtered media file in the information flow media scenario can be determined.

[0180] Step S410: Output the target value attribute by calling the second interface.

[0181] In the technical solution provided in step S410 of the present application, the second interface may include a second parameter, and the parameter value of the second parameter may include the target value attribute.

[0182] In this embodiment, after determining the target value attribute of the media file in the information flow media scenario based on the value attribute threshold, the target value attribute can be output by calling the second interface.

[0183] Through the above steps S402 to S410 of this application, the media data to be processed in the information flow media scenario is obtained by calling the first interface. The first interface includes a first parameter, and the parameter value of the first parameter includes the media data to be processed. The media data to be processed at least includes the media file to be delivered to the media delivery platform. The media data to be processed is input into the value prediction model, and a value distribution result is predicted. The value prediction model is trained using media data samples and the distribution results of value attribute samples of the media data samples. The value distribution result is used to represent the distribution results of different value attributes of the media file. Based on the value distribution result, a value attribute threshold of the media file is determined. The value attribute threshold is used to make the revenue information of the media delivery platform greater than the revenue information threshold. Based on the value attribute threshold, the target value attribute of the media file in the information flow media scenario is determined, where the target value attribute is greater than or equal to the value attribute threshold. The target value attribute is output by calling the second interface. The second interface includes a second parameter, and the parameter value of the second parameter includes the target value attribute. That is, in the information flow media scenario of this application embodiment, the value prediction model is used to predict the media data to be processed obtained by calling the first interface, and a value distribution result is obtained. Then, the value attribute threshold for the media data is calculated in real time using the value distribution result. The value attribute threshold is a personalized dynamic reserve price, rather than a traditional fixed reserve price. Finally, at the billing stage, it is ensured that the target value attribute is not lower than the value attribute threshold, and the target value attribute is output through the second interface, thereby achieving the goal of maximizing the revenue of the media delivery platform, and further realizing the technical effect of improving the flexibility of media data processing, and solving the technical problem of low flexibility in media data processing.

[0184] The embodiment of this application also provides a media data processing system. Figure 5 It is a schematic diagram of a media data processing system according to an embodiment of this application. As Figure 5 shown, the media data processing system 500 may include: a client 502 and a server 504.

[0185] The client 502 is used to transmit the media data to be processed in the information flow media scenario. The media data to be processed at least includes the media file to be delivered to the media delivery platform.

[0186] For example, the client 502 is used to transmit the media data to be processed in the scenario of streaming media in the information flow. The media data to be processed can be a <session, ad> pair, which can represent the display opportunity of an advertisement in a specific user session. The <session, ad> pair can include information about the user and their current behavior (from the session), as well as specific information about the advertisement to be displayed (from the advertisement). By analyzing the <session, ad> pair, the advertising system can more accurately evaluate the potential value of this display opportunity for the advertiser.

[0187] The server 504 is used to input the media data to be processed into the value prediction model to obtain the value distribution result. Among them, the value prediction model is trained using media data samples and the distribution results of the value attribute samples of the media data samples. The value distribution result is used to represent the distribution results of different value attributes of the media file; based on the value distribution result, the value attribute threshold of the media file is determined, where the value attribute threshold is used to make the revenue information of the media placement platform greater than the revenue information threshold; based on the value attribute threshold, the target value attribute of the media file in the scenario of streaming media in the information flow is determined, where the target value attribute is greater than or equal to the value attribute threshold; the target value attribute is output to the client.

[0188] For example, after obtaining the <session, ad> pair of the media data to be processed in the scenario of streaming media in the information flow transmitted by the client 502, the server 504 can be used to input the <session, ad> pair into the advertiser value distribution estimation model, and use the advertiser value distribution estimation model to estimate the advertiser value distribution probability density function f(v) for each <session, ad> pair, so as to obtain the value distribution result.

[0189] Optionally, after inputting the <session, ad> pair of the media data to be processed into the advertiser value distribution estimation model to obtain the value distribution result f(v), the server 504 can calculate the personalized reserve price r* according to the value distribution result f(v), and the calculated personalized reserve price r* can be used to make the revenue information of the advertising platform greater than the revenue information threshold, so as to meet the revenue maximization mechanism.

[0190] Optionally, after determining the value attribute threshold of the media file based on the value distribution result, based on the determined value attribute threshold, the server 504 can filter out the media files with bids lower than the value attribute threshold. Further based on the value attribute threshold, the target value attribute of the filtered media files in the scenario of streaming media in the information flow can be determined.

[0191] In the media data processing system 500, the media data to be processed in the in-stream media scenario is transmitted through the client 502. Among them, the media data to be processed at least includes the media files to be delivered to the media delivery platform. The media data to be processed is input into the value prediction model through the server 504, and a value distribution result is predicted. The value prediction model is trained using media data samples and the distribution results of value attribute samples of the media data samples. The value distribution result is used to represent the distribution results of different value attributes of the media files. Based on the value distribution result, a value attribute threshold of the media file is determined, where the value attribute threshold is used to make the revenue information of the media delivery platform greater than the revenue information threshold. Based on the value attribute threshold, the target value attribute of the media file in the in-stream media scenario is determined, where the target value attribute is greater than or equal to the value attribute threshold. The target value attribute is output to the client. That is to say, in the media data processing system 500 of the embodiment of the present application, the media data to be processed in the in-stream media scenario is transmitted to the server 504 through the client 502. The server uses the value prediction model to predict the transmitted media data to be processed to obtain a value distribution result, and then uses the value distribution result to calculate in real time the value attribute threshold for the media data. The value attribute threshold is a personalized dynamic reserve price, rather than a traditional fixed reserve price. Finally, at the billing stage, it is ensured that the target value attribute is not lower than the value attribute threshold, and the target value attribute is output to the client, so as to achieve the goal of maximizing the revenue of the media delivery platform, and further achieve the technical effect of improving the flexibility of media data processing, and solve the technical problem of low flexibility in media data processing.

[0192] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application, for example, the data for verification, are all information and data authorized by the user or fully authorized by all parties. And the collection, use, and processing of relevant data need to comply with the relevant laws, regulations, and standards of relevant countries and regions, and corresponding operation entrances are provided for users to choose to authorize or refuse.

[0193] Currently, the auction mechanism is the core module of the online advertising system. The goal of the auction mechanism is to maximize one or several goals (such as platform revenue, user experience, advertiser effect, etc.) under the conditions of meeting the mechanism properties, such as the incentive compatibility property IC of the mechanism and the individual rationality property (abbreviated as IR) of the bidders.

[0194] In an alternative example, the goal of the intelligent auction mechanism for multi-objective optimization can be abstracted as maximizing social welfare under multi-objective value metrics. The allocation rule can be abstracted as: greedy allocation or sequential decision-making allocation based on the comprehensive score (rankscore), and the charging rule can be abstracted as: second-price charging based on the minimum bid (critical_bid) for obtaining the current slot.

[0195] The essence of the above auction mechanism is to optimize the allocation efficiency, that is, the efficiency mechanism, and does not directly optimize the platform revenue, that is, the optimal mechanism. Classical auction theories (such as Myerson) provide auction mechanisms in the context of single-item and independent advertiser value distributions. However, the assumption of known advertiser value distributions in the theory does not match the real auction environment where the advertiser value distributions are unknown. Therefore, it is necessary to make a relatively accurate prediction of this distribution. In addition, it is also necessary to accurately characterize the advertiser behavior for more macro-control. For example, suppressing "lying flat merchants" and supporting "active merchants" to make the entire market competition more sufficient. The advertiser value distribution is the final manifestation of the overall behavior of advertiser behavior and bidding agents. Predicting the advertiser value distribution is a step under the goal of advertiser behavior modeling.

[0196] In an alternative example, the classical Myerson auction theory only gives the theoretical target solution under specific conditions and has no practical implementation plan for reference. The solution for the reserve price design of search advertising is implemented in keyword auctions and is inconsistent with the usage scenario of in-feed advertising. Moreover, the reserve price is calculated based on statistics and is not a dynamic reserve price. Therefore, at the overall solution level, there is currently no complete reserve price design solution for the in-feed advertising scenario.

[0197] To solve the above problems, this embodiment builds the ability to predict the advertiser value distribution and constructs an online link to apply the Myerson reserve price to the existing auction mechanism. This embodiment is the first practice of the Myerson auction theory in the in-feed advertising scenario and has achieved significant benefits. This embodiment proposes a data-driven advertiser value distribution prediction model to solve the problem of unknown advertiser value distributions in the Myerson auction theory. Based on the advertiser value distribution prediction at the traffic granularity, personalized dynamic reserve prices are calculated and applied in the existing auction mechanism link.

[0198] The above method of this embodiment will be further introduced below.

[0199] Figure 6It is a flowchart of an optional information flow advertisement auction according to an embodiment of the present application. As Figure 6 shown, it is divided according to whether there is a natural card scenario. First, in the case of no natural card scenario, the process of information flow advertisement auction may include the following steps:

[0200] Step S601, estimate the advertisements participating in the auction.

[0201] In the above step, the advertisements participating in the auction can be estimated, including estimating click-through rate, conversion rate, etc., to form an expected evaluation of the advertisement effect.

[0202] Step S602, place a bid according to the estimated data and its own goals and budget.

[0203] In the above step, the advertiser or its bid strategy agent places a bid according to the estimated data and its own goals and budget, and this bid will determine the competitiveness of the advertisement in the auction.

[0204] Step S603, auction the advertisement display opportunity or advertising space.

[0205] In the above step, the auction (Acution) usually adopts the Generalized Second Price Auction (abbreviated as GSP) mechanism to determine which advertisement will be displayed to the user.

[0206] Step S604, post-process the initially determined advertisement display order and allocation result of the auction.

[0207] In the above step, after the auction ends, based on post-processing strategies, such as filtering, reordering, etc., the initially determined advertisement display order and allocation result of the auction are post-processed to further optimize the efficiency and effect of advertisement display.

[0208] Step S605, calculate the final billing price of the advertisement.

[0209] In the above step, calculate the final billing price of the advertisement according to the GSP mechanism (i.e., GSP settlement).

[0210] That is to say, for the scenario without natural cards, after estimation (advertisement), bidding, Acution, post-processing strategy and GSP settlement, there is no fusion stage in the scenario without natural cards, and the precise competition stage directly determines the allocation result and billing result. Secondly, in the case of a natural card scenario, the process of information flow advertisement auction may include the following steps:

[0211] Step S610, estimate the fusion value of the advertisement and the recommended content.

[0212] In the above steps, in addition to estimating the advertisements, the value estimation of the recommended content is also added to make a more reasonable display decision between advertisements and non-advertisements (natural content).

[0213] Step S620, mix and sort the advertisements and the recommended content based on the fusion mixing and sorting mechanism.

[0214] In the above steps, the advertisements and the recommended content are mixed and sorted to form a unified list to balance the advertisement display and the user experience, ensuring that the content seen by the user includes both advertisements and natural content.

[0215] Step S630, perform settlement based on the fusion billing service.

[0216] In the above steps, in the scenario with natural cards, the unified generalized second price auction (abbreviated as uGSP) can be adopted for settlement to ensure the same billing logic for advertisements and natural content.

[0217] That is, for the scenario with natural cards, through the fusion value estimation (advertisement + recommendation), the fusion mixing and sorting mechanism, and the fusion billing service (uGSP settlement), the fusion stage determines the final allocation result and the billing result.

[0218] Figure 7 It is a flowchart of an information flow advertisement auction with a dynamic reserve price according to an embodiment of the present application. As Figure 7 shown, the process of the information flow advertisement auction with a dynamic reserve price in the scenario without natural cards may include the following steps:

[0219] Step S701, estimate the advertisements participating in the auction.

[0220] Step S702, place a bid according to the estimated data and its own goals and budget.

[0221] Step S703, auction the advertisement display opportunity or the advertisement space.

[0222] It should be noted that steps S701 to S703 are the same as steps S601 to S603, and will not be specifically described here.

[0223] Step S704, establish an advertisement owner value distribution estimation model.

[0224] In the above steps, based on the advertisement owner value distribution estimation model, the probability density function f(v) of the advertisement owner value distribution is estimated for each <session, ad> pair.

[0225] Step S705, calculate the personalized reserve price.

[0226] In the above steps, based on the probability density function f(v) of the advertiser value obtained by estimation, the personalized reserve price r* is calculated.

[0227] Step S706, determine whether to filter the advertisement according to the personalized reserve price.

[0228] In the above steps, it is determined whether the ad is filtered according to the bid of each advertisement ad and r*.

[0229] Step S707, post-process the preliminary advertisement display order and allocation result determined by the auction.

[0230] It should be noted that step S707 is the same as step S604, and will not be elaborated here.

[0231] Step S708, calculate the final billing price of the advertisement.

[0232] In the above steps, compared with step S605, step S708 updates the billing logic, that is, the settlement price not only considers the second-highest bid in GSP, but also compares it with the calculated personalized reserve price r*, and takes the larger value as the final billing price, thus ensuring that the billing price for the advertisement display is not lower than the platform's expectation of its value.

[0233] The process of the auction of the information flow advertisement with a dynamic reserve price in the natural card scenario may include the following steps:

[0234] Step S710, estimate the combined value of the advertisement and the recommended content.

[0235] Step S720, mix and sort the advertisement and the recommended content based on the combined mixing mechanism.

[0236] It should be noted that steps S710 to S720 are the same as steps S610 to S620, and will not be elaborated here.

[0237] Step S730, pass the personalized reserve price to the combined billing service.

[0238] In the above steps, a personalized reserve price field is added to the AIM return field to ensure that the combined billing service can obtain the personalized reserve price for billing.

[0239] Step S740, perform settlement based on the combined billing service.

[0240] In the above steps, compared with step S630, the billing service is updated, that is, the personalized reserve price r* is considered. When settling by uGSP, the system will compare the second-highest valid bid with r*, and take the larger one as the final billing price, while ensuring that this price is not lower than r* to protect the platform revenue.

[0241] It should be noted that the calculation logics of GSP and uGSP used by the settlement module (settlement in the refined competition stage and converged charging service) are the same. Since there is no re-bidding in the convergence stage, the ads entering the convergence stage meet the conditions. Therefore, there is no need for re-filtering in the convergence stage, and only the charging logic of the converged charging service needs to be changed.

[0242] The following further introduces the process of the above information flow ad auction with a dynamic reserve price in this embodiment.

[0243] Regarding the advertiser value distribution prediction model, different from traditional binary classification (such as CXR) prediction, the advertiser value distribution prediction is not a well-defined problem. The learning labels of traditional CXR prediction are objective facts, that is, real objective behaviors such as whether the user clicks or makes a deal, and it is assumed that the labels follow a Bernoulli distribution. However, the advertiser value is not an observable objective fact. The advertiser value is more of a latent variable in the advertiser's mind, which is not directly expressed or revealed to the platform. There is a theory that under the assumption of an advertiser who is a value maximizer, the GSP mechanism is IC. If a mechanism is IC, then telling the truth is a dominant strategy, that is, the bid is the advertiser's true valuation value for the current auction (subsequently represented by v). Based on the above assumptions, it can be considered that in the game equilibrium state of the current GSP mechanism, bid = v, and the label learned by the model is the bid.

[0244] Value is a continuous variable. Direct regression can only obtain a real value and cannot obtain a distribution, which is not convenient for subsequent use of the prediction results. If it is desired that the model output is a distribution, then an assumption needs to be made about the output form. First, what is actually to be predicted is the conditional probability p(v|x), that is, the distribution v of the advertiser value given the feature vector x. x can be expressed as {x1, x2,..., xn}, and x1 is a certain dimension feature of the feature vector x. However, the distribution form of p(v|x) for a given feature vector x is unknown. A sample of a certain traffic granularity is a sampling on this distribution, and the appearance of its distribution cannot be observed from the posterior data. Therefore, the frequency histogram of the posterior value v can be observed to approximately replace the probability distribution, that is, p(v), and it is observed that it basically follows a lognormal distribution. Further, by dividing and observing the frequency histogram of p(v) in different dimensions, what is actually observed is the conditional probability distributions of p(v|x1 = C1) and p(v|x1 = C2), and it is observed that they basically follow a lognormal distribution. Therefore, the distribution to be predicted can be assumed to be a lognormal distribution for formalization. After obtaining the mean, variance, input features, model parameters, etc. of the lognormal distribution predicted by the model, the loss function can be derived from the maximum likelihood estimation.

[0245] Figure 8 is a schematic structural diagram of an advertiser value distribution prediction model according to an embodiment of the present application. As Figure 8 shown, in terms of feature engineering, attribute features on the advertisement side (such as id features), pValue features on the advertisement side, pid features on the request side (i.e., context features), and statistical features on the advertisement side are selected. Since the information on the user side is already included in pValue, there is no need to repeat the modeling. The overall structure of the advertiser value distribution prediction model adopts a relatively simple Multilayer Perceptron (abbreviated as MLP) structure. On the one hand, it is not desired to make the model too heavy, increasing the inference pressure and resource increment; on the other hand, compared with complex model structures, effective features play a greater role. The pid feature on the request side and the id feature on the advertisement side are converted into dense vector representations through the Embedding Layer to capture the complex relationships and patterns within the features. It should be noted that, combined with some expert experience, in addition to directly adding the predicted pValue, feature combinations of different pValues are also performed to obtain other pre-estimated values with physical meanings. In addition, due to the understanding of a certain linear relationship with pValue, an element-wise multiplication layer is also introduced in the model structure, and an element-wise dot product model structure is performed on the embedding of the id feature and pValue. The pValue feature on the advertisement side and the statistical feature on the advertisement side are concatenated through the Concat layer to combine the feature information from different sources. After the concatenation layer, it is processed by the Dense Layer, which usually involves a fully connected neural network layer for non-linear transformation of the features. Finally, the mean and variance of the output distribution of the model are output.

[0246] Although the Myerson auction theory provides a mechanism for single-item auctions, in the complex auction environment of online advertising, heterogeneous multi-items, single-parameter, multi-round, and unconstrained conditions are still inevitable. Therefore, the mechanism for multi-item auctions remains an open issue. In an alternative example, guided by auction theory, experiments on reserve prices are conducted in search advertising, and conclusions consistent with the Myerson auction theory are obtained, that is, after introducing a new reserve price, the revenue increases. In addition, it is also proven that the GSP mechanism with a personalized reserve price r* is a revenue-maximizing mechanism, provided that the virtual valuation function is monotonically increasing and the buyer value distribution is independent and identically distributed (abbreviated as i.i.d). In this embodiment, the application of the Myerson reserve price includes three parts: calculating the personalized reserve price r* based on the estimation result of the advertiser value distribution; judging whether to filter competing ads according to the personalized reserve price r*; and the billing price in the billing stage should not be lower than the personalized reserve price r*.

[0247] According to the Myerson auction theory, based on the probability density function f(υ) of the lognormal distribution and the cumulative distribution function F(υ) of the lognormal distribution estimated from the advertiser value distribution, the virtual valuation function φ(υ) is calculated. Since the virtual valuation function corresponding to the lognormal distribution function satisfies the property of monotonic increase, the zero point of the virtual valuation function φ(υ) can be obtained by the bisection method, and this zero point is the personalized reserve price r*.

[0248] The personalized reserve price r* is the expected price of this traffic for competing ads considered by the platform. If the bid is lower than this price, it is considered that the competing ad's bid is too low and should lose the opportunity to participate in the competition. The most direct approach is to simply filter out the competing ads with bids lower than the personalized reserve price. However, complete filtering requires the platform to bear a greater risk of the auction not receiving any bids. Consider an extreme case: due to serious overestimation in model estimation, all the bids of competing ads for a certain request are lower than the personalized reserve price. After filtering, no ads are returned. If there is no fallback strategy at the front end, it will result in a blank screen for the user, with extremely high risk. Therefore, for stability considerations, this embodiment adopts a relatively conservative filtering strategy and sets a filtering ratio parameter filter_ratio. Let the queue length be N, and the number of ads filtered by the reserve price in the queue be M. If M / N > filter_ratio, the reserve price filtering logic does not take effect, the competing ads are retained, and the reserve price is set to 0; if M / N <= filter_ratio, then these M ads are filtered.

[0249] Currently, regardless of whether there is a fusion stage, the billing for each scenario is two-price billing (except for the full-site promotion), that is, the settlement price is the minimum bid to maintain the current slot. After introducing the personalized reserve price r*, the settlement price is settle_price = max(original critical_bid, r * ). The billing service is a module with relatively high risks. This embodiment can ensure that no more systematic risks (such as billing higher than the first price) are introduced after the reserve price is introduced compared to the original link. Due to the existence of the reserve price filtering logic, the r* of the advertisements entering the billing stage is strictly less than or equal to the first-price bid, so the deduction will not exceed the first price. If r * <= original critical_bid, it means that the reserve price does not bring an increase in revenue and is consistent with the existing system. If the original critical_bid < r * <= bid, then the reserve price is the source of revenue increase.

[0250] Compared with the classic Myerson auction theory, this embodiment implements and applies this theory to the information flow advertising scenario; compared with the statistical-based reserve price calculation method, this embodiment uses a data-driven approach to model the value distribution of advertisers and calculates the traffic granularity personalized reserve price based on this distribution.

[0251] This embodiment is a dynamic reserve price auction algorithm that implements the Myerson auction theory in the information flow advertising scenario, proposes a data-driven method to calculate the personalized dynamic reserve price of traffic granularity. After deploying the entire link in the information flow advertising scenario, through offline and online experiments, it is verified that the method in this embodiment can bring a significant increase in revenue to the platform.

[0252] In the embodiment of the present application, in the information flow media scenario, a value prediction model is used to predict the to-be-processed media data to obtain a value distribution result. Then, the value attribute threshold for the media data is calculated in real time using the value distribution result, and this value attribute threshold is a personalized dynamic reserve price rather than a traditional fixed reserve price. Finally, in the billing stage, it is ensured that the target value attribute is not lower than the value attribute threshold, thereby achieving the goal of maximizing the revenue of the media placement platform, and further achieving the technical effect of improving the flexibility of media data processing and solving the technical problem of low flexibility in media data processing.

[0253] Figure 9 It is a structural block diagram of the computing environment of a media data processing method according to the embodiment of the present application, as Figure 9As shown, the computing environment 901 includes multiple computing nodes (such as servers, shown in the figure as 910-1, 910-2, …) running on a distributed network. Each computing node contains local processing and memory resources, and end users 902 can remotely run applications or store data in the computing environment 901. Applications can be provided as multiple services 920-1, 920-2, 920-3, and 920-4 in the computing environment 901, representing services “A”, “D”, “E”, and “H” respectively.

[0254] End users 902 can provide and access services through a web browser or other software applications on the client. In some embodiments, the provisioning and / or requests of end users 902 can be provided to the ingress gateway 930. The ingress gateway 930 can include a corresponding proxy to handle the provisioning and / or requests for services (one or more services provided in the computing environment 901).

[0255] Services are provided or deployed according to various virtualization technologies supported by the computing environment 901. In some embodiments, services can be provided based on virtual machine (VM)-based virtualization, container-based virtualization, and / or similar means. VM-based virtualization can simulate a real computer by initializing a virtual machine and execute programs and applications without directly accessing any actual hardware resources. While virtualizing the machine with a virtual machine, according to container-based virtualization, containers can be launched to virtualize the entire operating system (OS) so that multiple workloads can run on a single operating system instance.

[0256] In one embodiment of container-based virtualization, several containers of a service can be assembled into a Pod (e.g., a Kubernetes Pod). For example, as Figure 9 shown, service 920-2 can be equipped with one or more Pods 940-1, 940-2, …, 940-N (collectively referred to as Pods). A Pod can include a proxy 945 and one or more containers 942-1, 942-2, …, 942-M (collectively referred to as containers). One or more containers in the Pod handle requests related to one or more corresponding functions of the service, and the proxy 945 generally controls network functions related to the service, such as routing, load balancing, etc. Other services can also be equipped with Pods similar to this one.

[0257] During operation, executing a user request from end user 902 may require invoking one or more services in the computing environment 901, and executing one or more functions of a service may require invoking one or more functions of another service. As Figure 9As shown, service "A" 920-1 receives a user request from end user 902 at ingress gateway 930. Service "A" 920-1 may invoke service "D" 920-2, and service "D" 920-2 may request service "E" 920-3 to perform one or more functions.

[0258] The computing environment described above may be a cloud computing environment where the allocation of resources is managed by a cloud service provider, allowing for the development of functions without considering the implementation, tuning, or scaling of servers. This computing environment allows developers to execute code in response to events without building or maintaining complex infrastructure. Services can be segmented into a set of functions that can automatically scale independently, rather than scaling a single hardware device to handle potential loads.

[0259] According to an embodiment of the present application, there is also provided a media data processing apparatus for implementing the above Figure 2 media data processing method for the media data shown.

[0260] Figure 10 is a schematic diagram of a media data processing apparatus according to an embodiment of the present application. As Figure 10 shown, the media data processing apparatus 1000 may include: a first acquisition unit 1002, a first prediction unit 1004, a first determination unit 1006, and a second determination unit 1008.

[0261] The first acquisition unit 1002 is configured to acquire media data to be processed in an in-stream media scenario, where the media data to be processed includes at least a media file to be delivered to a media delivery platform.

[0262] The first prediction unit 1004 is configured to input the media data to be processed into a value prediction model to obtain a value distribution result, where the value prediction model is trained using media data samples and the distribution results of value attribute samples of the media data samples, and the value distribution result is used to represent the distribution results of different value attributes of the media file.

[0263] The first determination unit 1006 is configured to determine a value attribute threshold for the media file based on the value distribution result, where the value attribute threshold is used to make the revenue information of the media delivery platform greater than the revenue information threshold.

[0264] The second determination unit 1008 is configured to determine a target value attribute of the media file in the in-stream media scenario based on the value attribute threshold, where the target value attribute is greater than or equal to the value attribute threshold.

[0265] It should be noted here that the above-mentioned first acquisition unit 1002, first prediction unit 1004, first determination unit 1006, and second determination unit 1008 correspond to steps S202 to S208. The implementation examples and application scenarios of the four units and the corresponding steps are the same, but are not limited to the content disclosed above. It should be noted that the above-mentioned modules or units can be hardware components or software components stored in the memory and processed by one or more processors. The above-mentioned modules can also be part of the device and can run in the server 10 provided in the above-mentioned embodiment.

[0266] In the media data processing device 1000, the first acquisition unit 1002 acquires the media data to be processed in the information flow media scenario. The media data to be processed includes at least the media file to be delivered to the media delivery platform. The first prediction unit 1004 inputs the media data to be processed into the value prediction model to obtain the value distribution result. The value prediction model is trained using the media data samples and the distribution results of the value attribute samples of the media data samples. The value distribution result is used to represent the distribution results of different value attributes of the media file. The first determination unit 1006 determines the value attribute threshold of the media file based on the value distribution result. The value attribute threshold is used to make the revenue information of the media delivery platform greater than the revenue information threshold. The second determination unit 1008 determines the target value attribute of the media file in the information flow media scenario based on the value attribute threshold. The target value attribute is greater than or equal to the value attribute threshold. That is to say, in the information flow media scenario, the embodiment of the present application uses the value prediction model to predict the media data to be processed, obtains the value distribution result, and then uses the value distribution result to calculate the value attribute threshold for the media data in real time. The value attribute threshold is a personalized dynamic reserve price, rather than the traditional fixed reserve price. Finally, in the billing stage, it is ensured that the target value attribute is not lower than the value attribute threshold, so as to achieve the goal of maximizing the revenue of the media delivery platform, and further achieve the technical effect of improving the flexibility of media data processing, and solve the technical problem of low flexibility of media data processing.

[0267] According to the embodiment of the present application, there is also provided a media data processing device for implementing the above Figure 3 shown media data processing method.

[0268] Figure 11 It is a schematic diagram of another media data processing device according to the embodiment of the present application. As Figure 11 shown, the media data processing device 1100 may include: a second acquisition unit 1102, a second prediction unit 1104, a third determination unit 1106, a fourth determination unit 1108, and a sending unit 1110.

[0269] A second acquisition unit 1102, configured to acquire media data to be processed in an information stream media scenario from an e-commerce platform, where the media data to be processed at least includes media files to be delivered to a media delivery platform.

[0270] A second prediction unit 1104, configured to input the media data to be processed into a value prediction model, and predict a value distribution result, where the value prediction model is trained using media data samples and distribution results of value attribute samples of the media data samples, and the value distribution result is used to represent distribution results of different value attributes of the media files.

[0271] A third determination unit 1106, configured to determine a value attribute threshold of the media file based on the value distribution result, where the value attribute threshold is used to make the revenue information of the media delivery platform greater than a revenue information threshold.

[0272] A fourth determination unit 1108, configured to determine a target value attribute of the media file in the information stream media scenario based on the value attribute threshold, where the target value attribute is greater than or equal to the value attribute threshold.

[0273] A sending unit 1110, configured to send the target value attribute to the e-commerce platform.

[0274] Here, the second acquisition unit 1102, the second prediction unit 1104, the third determination unit 1106, the fourth determination unit 1108, and the sending unit 1110 correspond to steps S302 to S310. The instances and application scenarios implemented by the five units and the corresponding steps are the same, but are not limited to the content disclosed above. It should be noted that the above modules or units may be hardware components or software components stored in a memory and processed by one or more processors, and the above modules may also be part of a device and may run in the server 10 provided in the above embodiment.

[0275] In the media data processing device 1100, the second acquisition unit 1102 acquires the media data to be processed in the information flow media scenario from an e-commerce platform. The media data to be processed at least includes media files to be delivered to a media delivery platform. The second prediction unit 1104 inputs the media data to be processed into a value prediction model to obtain a value distribution result. The value prediction model is trained using media data samples and the distribution results of value attribute samples of the media data samples. The value distribution result is used to represent the distribution results of different value attributes of the media files. The third determination unit 1106 determines a value attribute threshold for the media files based on the value distribution result. The value attribute threshold is used to make the revenue information of the media delivery platform greater than the revenue information threshold. The fourth determination unit 1108 determines the target value attribute of the media files in the information flow media scenario based on the value attribute threshold. The target value attribute is greater than or equal to the value attribute threshold. The sending unit 1110 sends the target value attribute to the e-commerce platform. That is to say, in the information flow media scenario, the embodiment of the present application uses a value prediction model to predict the media data to be processed acquired from an e-commerce platform to obtain a value distribution result, and then uses the value distribution result to calculate in real time the value attribute threshold for the media data. The value attribute threshold is a personalized dynamic reserve price, rather than a traditional fixed reserve price. Finally, in the billing stage, it is ensured that the target value attribute is not lower than the value attribute threshold, thereby achieving the goal of maximizing the revenue of the media delivery platform, and further realizing the technical effect of improving the flexibility of media data processing and solving the technical problem of low flexibility in media data processing.

[0276] According to an embodiment of the present application, there is also provided a media data processing device for implementing the above Figure 4 shown media data processing method.

[0277] Figure 12 is a schematic diagram of another media data processing device according to an embodiment of the present application, as Figure 12 shown. The media data processing device 1200 may include: a third acquisition unit 1202, a third prediction unit 1204, a fifth determination unit 1206, a sixth determination unit 1208, and an output unit 1210.

[0278] The third acquisition unit 1202 is configured to acquire the media data to be processed in the information flow media scenario by invoking a first interface. The first interface includes a first parameter, and the parameter value of the first parameter includes the media data to be processed. The media data to be processed at least includes media files to be delivered to a media delivery platform.

[0279] A third prediction unit 1204 is configured to input media data to be processed into a value prediction model, and predict a value distribution result, where the value prediction model is trained by using media data samples and distribution results of value attribute samples of the media data samples, and the value distribution result is used to represent distribution results of different value attributes of a media file.

[0280] A fifth determination unit 1206 is configured to determine a value attribute threshold of a media file based on the value distribution result, where the value attribute threshold is used to make the revenue information of a media placement platform greater than a revenue information threshold.

[0281] A sixth determination unit 1208 is configured to determine a target value attribute of a media file in an in-stream media scenario based on the value attribute threshold, where the target value attribute is greater than or equal to the value attribute threshold.

[0282] An output unit 1210 is configured to output the target value attribute by invoking a second interface, where the second interface includes a second parameter, and the parameter value of the second parameter includes the target value attribute.

[0283] It should be noted here that the above-mentioned third acquisition unit 1202, third prediction unit 1204, fifth determination unit 1206, sixth determination unit 1208, and output unit 1210 have the same implemented examples and application scenarios as the corresponding steps, but are not limited to the above-disclosed content. It should be noted that the above modules or units may be hardware components or software components stored in a memory and processed by one or more processors, and the above modules may also be part of a device and may run in the server 10 provided in the above embodiment.

[0284] In the processing device 1200 for media data, a third acquisition unit 1202 acquires media data to be processed in the information flow media scenario by invoking a first interface. The first interface includes a first parameter, and the parameter value of the first parameter includes the media data to be processed. The media data to be processed at least includes media files to be delivered to a media delivery platform. A third prediction unit 1204 inputs the media data to be processed into a value prediction model to obtain a value distribution result. The value prediction model is trained using media data samples and the distribution results of value attribute samples of the media data samples. The value distribution result is used to represent the distribution results of different value attributes of the media files. A fifth determination unit 1206 determines a value attribute threshold for the media files based on the value distribution result. The value attribute threshold is used to make the revenue information of the media delivery platform greater than a revenue information threshold. A sixth determination unit 1208 determines a target value attribute of the media files in the information flow media scenario based on the value attribute threshold. The target value attribute is greater than or equal to the value attribute threshold. An output unit 1210 outputs the target value attribute by invoking a second interface. The second interface includes a second parameter, and the parameter value of the second parameter includes the target value attribute. That is to say, in the information flow media scenario, the embodiment of the present application uses a value prediction model to predict the media data to be processed acquired by invoking the first interface, obtains a value distribution result, and then uses the value distribution result to calculate in real time the value attribute threshold for the media data. The value attribute threshold is a personalized dynamic reserve price rather than a traditional fixed reserve price. Finally, in the billing stage, it is ensured that the target value attribute is not lower than the value attribute threshold, and the target value attribute is output through the second interface, thereby achieving the goal of maximizing the revenue of the media delivery platform, and further achieving the technical effect of improving the flexibility of media data processing, and solving the technical problem of low flexibility in media data processing.

[0285] It should be noted that the preferred implementation schemes involved in the above embodiments of the present application are the same as the schemes, application scenarios, and implementation processes provided in the above embodiments, but are not limited to the schemes provided in the above embodiments.

[0286] An embodiment of the present application can provide a computing device. Figure 13 is a structural block diagram of a computing device according to an embodiment of the present application, as Figure 13 shown. The computing device 1300 may include: one or more (only one is shown in the figure) processors 1302, a memory 1304, a storage controller, and a peripheral interface.

[0287] The above-mentioned computing device can be understood as an integrated intelligent terminal, including but not limited to servers, desktop computers, personal computers (abbreviated as PC), model all-in-ones, etc. Moreover, the above-mentioned models described in the embodiments of the present application can be pre-installed in the computing device.

[0288] Specifically, the computing device can pre-install various types of models, including but not limited to models in the fields of natural language processing, visual processing, speech processing, code processing, multi-modal task processing, etc., so as to provide diverse model selections. In different product forms, the computing device can support one or more model usage methods, including but not limited to model training, model invocation, model fine-tuning, model deployment, model inference and application, etc. In some product forms, the computing device also supports model management, including but not limited to multi-type model management (supporting the management of various types of models such as discriminative and generative), model version control (supporting the control of different model versions), model evaluation (evaluating the performance and effect of the model based on model evaluation tools), etc. In other product forms, the computing device can also create applications based on the model, provide the ability to call the Application Programming Interface (abbreviated as API), and can call the model into the created application through the API interface, while providing application management tools to realize the control of the application.

[0289] Furthermore, the computing device can also include data management (supporting the creation and management of model tuning data sets), a training center (providing rich training resources to help users learn and master AI technologies), and basic control capabilities (providing enterprise-level basic control capabilities to ensure the security and efficient operation of the system). Through the above functions, a comprehensive and integrated AI development, training, deployment, and application device is provided.

[0290] Among them, the memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the methods and devices in the embodiments of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, to implement the methods in the above embodiments. The memory can include high-speed random access memory, and can also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memories. In some instances, the memory can further include memories remotely set relative to the processor, and these remote memories can be connected to terminal A through a network. Examples of the above network include but are not limited to the Internet, enterprise intranets, local area networks, mobile communication networks, and their combinations.

[0291] The processor can call the executable program stored in the memory through a transmission device to execute the method described in any one of the above embodiments.

[0292] Embodiments of the present application can provide an electronic device. Figure 14 is a structural block diagram of an electronic device according to an embodiment of the present application, as Figure 14 shown, the electronic device may include: an input / output device 1402; a memory 1404 and a processor 1406, wherein the processor 1406 is connected to the input / output device 1402 and the memory 1404 through a bus 1408.

[0293] Among them, the memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the methods and devices in the embodiments of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, implements the methods in the above embodiments. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some instances, the memory may further include a memory remotely provided relative to the processor, and these remote memories can be connected to the terminal A through a network. Examples of the above network include but are not limited to the Internet, enterprise intranets, local area networks, mobile communication networks, and combinations thereof.

[0294] The processor can call the executable program stored in the memory through a transmission device to execute the method described in any one of the above embodiments.

[0295] Those of ordinary skill in the art can understand that the structure shown in the figure is only schematic, and the computing device can also be a terminal device such as a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a handheld computer, and a Mobile Internet Device (MID), a PAD, etc. The figure does not limit the structure of the above computing device. For example, the computing device 1000 may further include more or fewer components (such as a network interface, a display device, etc.) than those shown in the figure, or have a different configuration from that shown in the figure.

[0296] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the relevant hardware of the terminal device through a program, and the program can be stored in a computer-readable storage medium. The storage medium may include: a flash drive, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disc, etc.

[0297] Embodiments of the present application also provide a computer-readable storage medium. Optionally, in this embodiment, the above computer-readable storage medium may be used to store the program code executed by the method provided in the above embodiment.

[0298] Optionally, in this embodiment, the above storage medium may be located in a computing device.

[0299] Optionally, in this embodiment, the computer-readable storage medium is configured to store an executable program, and when the executable program runs, it controls the device where the computer-readable storage medium is located to execute the method described in any one of the above embodiments.

[0300] Embodiments of the present application also provide a computer program product. Optionally, in this embodiment, the above computer program product may include a computer program, and the above computer program implements the method provided in the above embodiment when executed by a processor.

[0301] Embodiments of the present application also provide a computer program product. Optionally, the above computer program product may include a non-volatile computer-readable storage medium, and the above non-volatile computer-readable storage medium may be used to store a computer program, and the above computer program implements the method provided in the above embodiment when executed by a processor.

[0302] Embodiments of the present application also provide a computer program. Optionally, in this embodiment, the above computer program implements the method provided in the above embodiment when executed by a processor.

[0303] In the above embodiments of the present application, the descriptions of the respective embodiments have their own emphases. For parts not detailed in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0304] In several embodiments provided by the present application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection between each other can be through some interfaces. The indirect coupling or communication connection of units or modules can be in an electrical or other form.

[0305] The unit described as a separation component may or may not be physically separated. The component shown as a unit may or may not be a physical unit, that is, it may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0306] In addition, each functional unit in various embodiments of the present application may be integrated in a processing unit, may exist separately as individual physical units, or two or more units may be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0307] If the above-mentioned integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present application. The foregoing storage medium includes: various media such as USB flash drives, read-only memory ROM, random access memory RAM, mobile hard disks, magnetic disks or optical discs that can store program codes.

[0308] The above is only the preferred embodiment of the present application. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. A method for processing media data, characterized in that: include: Acquire the media data to be processed in the information streaming media scenario, wherein the media data to be processed at least includes a media file to be delivered to the media delivery platform; Inputting the to-be-processed media data into a value prediction model to predict a value distribution result, wherein the value prediction model is trained using a media data sample and a distribution result of a value attribute sample of the media data sample, and the value distribution result is used to represent a distribution result of different value attributes of the media file; Based on the value distribution result, determining a value attribute threshold of the media file, wherein the value attribute threshold is used to make the revenue information of the media delivery platform greater than a revenue information threshold; Based on the value attribute threshold, a target value attribute of the media file in the information streaming media scenario is determined, wherein the target value attribute is greater than or equal to the value attribute threshold.

2. The method according to claim 1, characterized in that The value distribution result includes the distribution result of the transaction probability of the media file under different value attributes. The transaction probability is used to indicate the possibility of the media delivery object performing a transaction operation for the media file according to the corresponding value attribute. The operation result corresponding to the transaction operation is used to allow the media delivery object to deliver the media file on the media delivery platform.

3. The method according to claim 1, characterized in that Determining a value attribute threshold of the media file based on the value distribution result includes: Converting the value distribution result into a virtual valuation function, wherein the virtual valuation function satisfies a monotonic change rule; Based on the virtual valuation function, the value attribute threshold is determined.

4. The method according to claim 3, characterized in that The value distribution result is described by a probability density function in a normal distribution state, and the value distribution result is converted into a virtual valuation function, including: Determine a cumulative distribution function corresponding to the probability density function, wherein the cumulative distribution function is used to represent the cumulative result of the distribution results of different value attributes, and the cumulative distribution function is in the normal distribution state; The cumulative distribution function and the probability density function are converted into the virtual valuation function.

5. The method according to claim 3, characterized in that: Determining the value attribute threshold based on the virtual valuation function includes: Determining an input value for making an output result of the virtual valuation function zero; The input value is determined as the value attribute threshold.

6. The method according to claim 1, characterized in that The to-be-processed media data is input into a value prediction model to predict a value distribution result, including: Extracting feature data from the to-be-processed media data, wherein the feature data is used to represent the properties of the media file and / or the properties of a media delivery object, and the media delivery object is used to deliver the media file to the media delivery platform; The characteristic data is input into the value prediction model, and the mean and variance in the value prediction model are used to convert the characteristic data into the value distribution result that satisfies the normal distribution state, wherein the mean and variance match the normal distribution state.

7. The method according to claim 6, characterized in that The to-be-processed media data samples at least include media file samples to be delivered to the media delivery platform, and the method further includes: Extracting feature data samples from the media data samples, wherein the feature data samples are used to represent properties of the media file samples and / or properties of the media delivery object samples, and the media delivery object samples are used to deliver the media file samples to the media delivery platform; Respectively determine distribution result samples of different value attribute samples of the media file sample under the feature data of different dimensions in the feature data sample; In response to the distribution result sample being in a normal distribution state, determining the mean and the variance matching the normal distribution state; The value prediction model is trained using the mean and the variance.

8. The method according to claim 7, characterized in that The method further comprises: Determining different operation behavior data corresponding to the media file sample in the feature data sample, wherein the operation behavior data is used to represent the operation behavior acting on the media file sample; Combining different operation behavior data to obtain a combination result; In response to the distribution result sample being in a normal distribution state, determining the mean and the variance that match the normal distribution state, including: in response to the distribution result sample being in the normal distribution state, using the combination result, determining the mean and the variance that match the normal distribution state.

9. The method according to claim 7, characterized in that: The value prediction model is obtained by training using the mean and the variance, including: Using the mean and the variance, a loss function is established; Using the loss function, determining the model parameters of the value prediction model to be generated; The value prediction model including the mean, the variance and the model parameters is established.

10. The method according to claim 1, characterized in that Determining the target value attribute of the media file in the information streaming media scenario based on the value attribute threshold includes: Determining a first initial value attribute of the media file in the information streaming media scenario; Determining a maximum value attribute between the first initial value attribute and the value attribute threshold; The value attribute that is greater than or equal to the maximum value attribute is determined as the target value attribute.

11. The method according to any one of claims 1 to 10, characterized in that The method further comprises: Acquire a second initial value attribute of the media file fed back by the media delivery platform; Based on the value attribute threshold, determining the target value attribute of the media file in the information streaming media scenario, including: when the information streaming media scenario is a first type of information streaming media scenario, in response to the second initial value attribute being greater than or equal to the value attribute threshold, determining the target value attribute of the media file in the information streaming media scenario based on the value attribute threshold, wherein the first type of information streaming media scenario is used to represent an information streaming media scenario that includes media content of the media file but does not include natural content.

12. The method according to claim 11, characterized in that The media file is from a media file queue, and the method further includes: In the media file queue, determining the number of to-be-filtered media files whose second initial value attribute is less than the value attribute threshold; Determining a ratio between the number of the to-be-filtered media files and the number of media files in the media file queue; In a case where the ratio is less than or equal to a ratio threshold, in response to the second initial value attribute being less than the value attribute threshold, the to-be-filtered media file is filtered out from the media file queue.

13. The method according to claim 12, characterized in that The method further comprises: When the ratio is greater than the ratio threshold, retaining the to-be-filtered media file in the media file queue; The value attribute of the to-be-filtered media file is set as a default value attribute.

14. The method according to any one of claims 1 to 10, characterized in that Determining the target value attribute of the media file in the information streaming media scenario based on the value attribute threshold includes: In the case where the information streaming media scenario is a second type of information streaming media scenario, transparently transmitting the value attribute threshold to the converged billing service of the media delivery platform, wherein the second type of information streaming media scenario is used to represent the media content including the media file and includes an information streaming media scenario of natural content; In the convergent billing service, the target value attribute of the media file in the information streaming media scenario is determined based on the value attribute threshold.

15. The method according to any one of claims 1 to 10, characterized in that Obtain the media data to be processed in the information streaming media scenario, including: In response to the media delivery platform generating traffic, obtaining the media data to be processed in the information streaming media scenario; Determining a value attribute threshold of the media file based on the value distribution result includes: determining the value attribute threshold of the traffic for the media file based on the value distribution result.

16. A method for processing media data, characterized in that: include: Acquire the media data to be processed in the information streaming media scenario from the e-commerce platform, wherein the media data to be processed at least includes the media files to be delivered to the media delivery platform; Inputting the to-be-processed media data into a value prediction model to predict a value distribution result, wherein the value prediction model is trained using a media data sample and a distribution result of a value attribute sample of the media data sample, and the value distribution result is used to represent a distribution result of different value attributes of the media file; Based on the value distribution result, determining a value attribute threshold of the media file, wherein the value attribute threshold is used to make the revenue information of the media delivery platform greater than a revenue information threshold; Based on the value attribute threshold, determining a target value attribute of the media file in the information streaming media scenario, wherein the target value attribute is greater than or equal to the value attribute threshold; The target value attribute is sent to the e-commerce platform.

17. The method according to claim 16, characterized in that Determining a value attribute threshold of the media file based on the value distribution result includes: Converting the value distribution result into a virtual valuation function, wherein the virtual valuation function satisfies a monotonic change rule; Determining the value attribute threshold based on the virtual valuation function; The method also includes: sending the value attribute threshold to the e-commerce platform.

18. A method for processing media data, characterized in that: include: Acquire the media data to be processed in the information streaming media scenario by calling a first interface, wherein the first interface includes a first parameter, a parameter value of the first parameter includes the media data to be processed, and the media data to be processed includes at least a media file to be delivered to a media delivery platform; Inputting the to-be-processed media data into a value prediction model to predict a value distribution result, wherein the value prediction model is trained using a media data sample and a distribution result of a value attribute sample of the media data sample, and the value distribution result is used to represent a distribution result of different value attributes of the media file; Based on the value distribution result, determining a value attribute threshold of the media file, wherein the value attribute threshold is used to make the revenue information of the media delivery platform greater than a revenue information threshold; Based on the value attribute threshold, determining a target value attribute of the media file in the information streaming media scenario, wherein the target value attribute is greater than or equal to the value attribute threshold; The target value attribute is output by calling a second interface, wherein the second interface includes a second parameter, and a parameter value of the second parameter includes the target value attribute.

19. A media data processing system, characterized in that: include: A client, used for transmitting the media data to be processed in the information streaming media scenario, wherein the media data to be processed at least includes a media file to be delivered to the media delivery platform; The server is used to input the to-be-processed media data into a value prediction model to predict a value distribution result, wherein the value prediction model is trained using media data samples and distribution results of value attribute samples of the media data samples, and the value distribution result is used to represent the distribution results of different value attributes of the media file; based on the value distribution result, a value attribute threshold of the media file is determined, wherein the value attribute threshold is used to make the revenue information of the media delivery platform greater than a revenue information threshold; based on the value attribute threshold, a target value attribute of the media file in the information streaming media scenario is determined, wherein the target value attribute is greater than or equal to the value attribute threshold; and the target value attribute is output to the client.

20. A computing device, characterized in that include: A memory storing an executable program; A processor, configured to run the program, wherein the program executes the method according to any one of claims 1 to 18 when running.

21. An electronic device, characterized in that: include: A memory storing an executable program; A processor, connected to the memory via a bus, and configured to run the program, wherein the program executes the method described in any one of claims 1 to 18 when running.

22. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored executable program, wherein when the executable program is executed, the device where the computer-readable storage medium is located is controlled to execute the method according to any one of claims 1 to 18.

23. A computer program product, characterized in that The invention comprises a computer program which, when executed by a processor, implements the method according to any one of claims 1 to 18.

Images