Content delivery method and apparatus, server, and storage medium

Abstract

The application discloses a content delivery method and device, a server and a storage medium. The method comprises the following steps: determining a target user amount of target content in a next target time period; determining target users associated with the target content in the next target time period according to the target user amount and a target user determination strategy; screening the target users associated with the target content in the next target time period to obtain to-be-delivered users, and sending the target content to the to-be-delivered users; at the end of the next target time period, updating a current real exposure amount of the target content at the end of a previous target time period by using a real exposure amount of the target content in the to-be-delivered users, returning to the first step until the real exposure amount at the end of the next target time period accumulates to reach a target exposure amount or until a corresponding accumulated time length at the end of the next target time period reaches a target time length. The method can reduce the over-exposure amount in the content delivery process.

Description

Technical Field

[0001] This application relates to the field of Internet information processing technology, and more specifically, to a content delivery method, apparatus, server, and storage medium. Background Technology

[0002] With the rapid development of mobile internet and information flow products, users can access and obtain a wealth of diverse information and content. In information flow scenarios, it is usually necessary to target various types of information and content to corresponding users, i.e., content delivery.

[0003] In the content delivery methods of related technologies, when delivering content with a small demand for exposure, real-time exposure data is used to control the content delivery process. However, due to the delay in exposure statistics, there is a problem of over-exposure when delivering content. Furthermore, since there is a large amount of content to be delivered in the delivery system, the over-exposure of exposure for a single content delivery task will lead to a large over-exposure of exposure in the entire delivery system. Summary of the Invention

[0004] In view of the above problems, this application proposes a content delivery method, apparatus, server and storage medium to improve the above problems.

[0005] In a first aspect, embodiments of this application provide a content delivery method, the method comprising: determining the target user volume of the target content in a subsequent target time period based on the target exposure volume, screening rate, and current actual exposure volume at the end of the previous target time period, wherein the previous target time period and the subsequent target time period are two adjacent time periods; determining the target users associated with the target content in the subsequent target time period according to the target user volume and target user determination strategy; screening the target users associated with the target content in the subsequent target time period to obtain users to be delivered to, and sending the target content to the users to be delivered to; at the end of the subsequent target time period, updating the current actual exposure volume of the target content at the end of the previous target time period based on the actual exposure volume, screening rate, and current actual exposure volume at the end of the previous target time period, and returning to the step: determining the target user volume of the target content in the subsequent target time period based on the target exposure volume, screening rate, and current actual exposure volume at the end of the previous target time period, until the cumulative actual exposure volume at the end of the subsequent target time period reaches the target exposure volume or until the cumulative duration at the end of the subsequent target time period reaches the target duration.

[0006] Secondly, embodiments of this application provide a content delivery device, which includes: a target user volume determination module, a target user determination module, a filtering and sending module, and a loop execution module. The system comprises the following modules: a target user volume determination module, which determines the target user volume of the target content in the next target time period based on the target exposure volume, filtering rate, and the current actual exposure volume at the end of the previous target time period (where the previous and next target time periods are adjacent); a target user determination module, which determines the target users associated with the target content in the next target time period based on the target user volume and target user determination strategy; a filtering and sending module, which filters the target users associated with the target content in the next target time period to obtain the users to be targeted to, and sends the target content to the users to be targeted to; and a loop execution module, which updates the current actual exposure volume of the target content at the end of the previous target time period using the actual exposure volume of the target content among the users to be targeted to at the end of the next target time period, and returns to the step: determining the target user volume of the target content in the next target time period based on the target exposure volume, filtering rate, and current actual exposure volume at the end of the previous target time period, until the cumulative actual exposure volume at the end of the next target time period reaches the target exposure volume or until the cumulative duration at the end of the next target time period reaches the target duration.

[0007] Optionally, the target user determination module includes: an associated user determination submodule and a target user determination submodule.

[0008] The associated user determination submodule is used to determine associated users from content request users based on the first feature information of the target content and the second feature information of the content request user.

[0009] The target user determination submodule is used to identify the previous target user group and associated users as the target users associated with the target content in the next target time period.

[0010] Optionally, the first feature information includes first attribute information and first historical behavior information of the user who clicked the target content, and the second feature information includes second attribute information and second historical behavior information of the content requesting user. The associated user determination submodule is further configured to determine the content requesting user as a candidate user when the second attribute information matches the first attribute information; perform feature extraction on the second historical behavior information of the candidate user to obtain a second feature vector corresponding to the candidate user, and perform feature extraction on the first historical behavior information of the user who clicked the target content to obtain a first feature vector; and determine the associated user from the candidate users based on the similarity between the first feature vector and the second feature vector and a similarity threshold.

[0011] Optionally, the loop execution module is also used to update the similarity threshold at the end of the next target time period.

[0012] Optionally, the target content is one of a preset number of candidate contents ranked first in similarity among multiple candidate contents corresponding to the associated user, and the candidate content is the content to be delivered whose similarity meets the similarity threshold.

[0013] Optionally, the target content is cold start content, and the associated user determination submodule is also used to obtain second attribute information that matches the first attribute information of the target content; and to obtain a target number of historical users with the second attribute information as users who clicked on the target content.

[0014] Optionally, the associated user determination submodule is also used to record users who clicked on the target content among the users to be targeted, and to update the users who clicked on the target content using the users who clicked on the target content.

[0015] Optionally, the first attribute information is obtained through the following steps: obtaining descriptive information corresponding to the target content; performing attribute prediction on the descriptive information through a neural network model to obtain the first attribute information output by the neural network model, wherein the neural network model is trained by descriptive information carrying attribute information labels.

[0016] Optionally, the associated user determination submodule is also used to obtain the active time corresponding to the content requesting user; determine candidate associated users from the content requesting users whose active time does not meet the time threshold; and determine the associated user from the candidate associated users based on the first feature information of the target content and the second feature information of the candidate associated users.

[0017] Optionally, the current actual exposure at the end of the previous target time period is the cumulative actual exposure at the end of the previous target time period. The loop execution module is also used to take the cumulative actual exposure of the target content among the users to be delivered at the end of the next target time period and the current actual exposure as the updated current actual exposure at the end of the previous target time period.

[0018] Optionally, the target user volume determination module is also used to acquire historical exposure distribution data, which characterizes the distribution of historical exposure over time; based on the historical exposure distribution data and the target exposure, determine the sub-target exposure of the target content in each target time period; based on the sub-target exposure in the previous target time period and the current actual exposure, determine the exposure error, and the current actual exposure at the end of the previous target time period is the actual exposure in the previous target time period; based on the sub-target exposure, exposure error, and filtering rate of the target content in the next target time period, determine the target user volume of the target content in the next target time period. Correspondingly, the loop execution module is also used to, at the end of the next target time period, use the actual exposure of the target content among the users to be targeted as the current actual exposure at the end of the previous target time period corresponding to the target content.

[0019] Optionally, the device also includes a target exposure determination module, used to determine the target exposure for the target content based on the exposure exchange information and the conversion ratio corresponding to the exposure exchange information.

[0020] Thirdly, embodiments of this application provide a server, including a processor and a memory; one or more programs are stored in the memory and configured to be executed by the processor to implement the above-described method.

[0021] Fourthly, embodiments of this application provide a computer-readable storage medium storing program code, wherein the above-described method is executed when the program code is run by a processor.

[0022] Fifthly, embodiments of this application provide a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the method described above.

[0023] This application provides a content delivery method, apparatus, server, and storage medium. By controlling the number of target users associated with target content within a subsequent target time period, the method avoids continuous content delivery, reduces the total amount of content delivered, and thus reduces over-exposure. Furthermore, by using the time interval of the target time period to count the actual exposure, the method effectively waits for the actual exposure, mitigating the impact of delays and further reducing over-exposure. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This paper shows a schematic diagram of the architecture of a content delivery system proposed in an embodiment of this application;

[0026] Figure 2 This illustration shows a schematic diagram of a process for displaying an information flow display interface according to an embodiment of this application;

[0027] Figure 3 This illustration shows a schematic diagram of another process for displaying an information flow display interface according to an embodiment of this application;

[0028] Figure 4 A flowchart of a content delivery method proposed in an embodiment of this application is shown;

[0029] Figure 5 This document shows a flowchart of one implementation of S110 in a content delivery method according to an embodiment of this application.

[0030] Figure 6 A flowchart of another content delivery method proposed in an embodiment of this application is shown;

[0031] Figure 7 This document shows a flowchart of one implementation of S220 in a content delivery method according to an embodiment of this application;

[0032] Figure 8 A flowchart of another content delivery method proposed in an embodiment of this application is shown;

[0033] Figure 9 This paper presents a flowchart illustrating a process for obtaining first attribute information according to an embodiment of this application.

[0034] Figure 10 This document illustrates a flowchart of an embodiment of the present application for obtaining first historical behavior information of a user who clicked on target content;

[0035] Figure 11 A flowchart of another content delivery method proposed in an embodiment of this application is shown;

[0036] Figure 12 A block diagram of a content delivery device according to an embodiment of this application is shown;

[0037] Figure 13A structural block diagram of a server for performing a content delivery method according to an embodiment of this application is shown. Detailed Implementation

[0038] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0039] In the following description, the terms "first" and "second" are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first" and "second" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0040] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0041] It should be noted that the content to be delivered in the content delivery method disclosed in this application embodiment, i.e., the target content, or the first feature information of the target content, can be stored on the blockchain.

[0042] Before providing a further detailed description of the embodiments of this application, some of the nouns and terms involved in the embodiments of this application are explained. The nouns and terms involved in the embodiments of this application shall be interpreted as follows:

[0043] 1) Content: Generally consists of a title and body text. The title can consist of a text title and a cover image, guiding users to click and read the body text; the body text can consist of text, images, videos, etc., to comprehensively display the information in the content.

[0044] 2) Information flow: refers to a scrollable information flow composed of multiple pieces of content. In the information flow, there are multiple content display sub-areas, which are arranged sequentially. Each content display sub-area displays the title of a piece of content. After the user clicks on the title, they can jump to the main content page.

[0045] 3) Exposure: When a piece of content is displayed to a user in the information feed, each time the content appears in the information feed, it is counted as one exposure, i.e., one exposure.

[0046] 4) Click count: When a piece of content is exposed to a user in the news feed, and the user clicks on the title to enter the main text page to read it, it is counted that the content has been clicked once, and a click count is recorded.

[0047] 5) Screening rate: Also known as funnel rate. After the ad delivery backend initially determines that a certain content can be delivered to a certain user, it will return the content and associated users to the upstream. The upstream various screening backends screen the content through set weighting, scoring, sorting, content diversity control strategies, etc., to determine the content that actually needs to be sent to the user's client. Therefore, there is a screening rate between the actual exposure volume of the client and the return volume initially determined by the ad delivery backend.

[0048] With the rapid development of mobile internet and information feed products, users can access and obtain a wealth of diverse information and content. Information feed products typically use machine learning to train recommendation models based on user behavior and interests on the platform, establishing a recommendation system to recommend other content that users may be interested in.

[0049] However, recommendation models and results, in certain specific scenarios, require intervention through delivery methods to influence the recommendation outcomes. Therefore, news feed products typically also establish a content delivery system, using manually selected conditions to target various types of information and content to corresponding users—this is content delivery.

[0050] However, the inventors discovered that in the content delivery methods of related technologies, when delivering content with low exposure requirements, real-time exposure data is directly used to control the content delivery process. However, on the one hand, the user's reported data and the calculation service are asynchronous, which is near real-time, with a time difference of several minutes; on the other hand, the delivery service updates the list of all delivery tasks on a timed basis, every few minutes, which also has a time difference; furthermore, each user refresh may contain more than 10 pieces of content, but only a portion of the content is displayed on the terminal screen each time. If the content to be delivered is not currently displayed, it will only be counted in the real-time exposure count when the user scrolls down and the subsequent content is displayed, which also creates a time difference. Therefore, under the influence of the above-mentioned time differences, there is a delay in exposure statistics. That is, after the real-time exposure count reaches the target exposure count and content delivery stops, there may still be exposures, which leads to the problem of over-exposure when delivering content. That is, the actual exposure count generated is greater than the set target exposure count. Moreover, since there is a large amount of content to be delivered in the delivery system, the over-exposure of exposure for a single content delivery task will lead to a large over-exposure of exposure in the entire delivery system.

[0051] To address the aforementioned issues, the inventors have proposed the content delivery method, apparatus, server, and storage medium provided in this application. By controlling the number of target users within a subsequent target time period, the target content can be associated with target users within that time period. Since target users are controlled, continuous content delivery can be avoided, reducing the total amount of content delivered and thus reducing over-exposure. Furthermore, by using the time interval of the target time period to count the actual exposure, the acquisition of the actual exposure has a certain waiting effect, which can mitigate the impact of delay and further reduce over-exposure.

[0052] Before providing a more detailed description of the embodiments of this application, an application environment related to the embodiments of this application will be introduced.

[0053] See Figure 1 , Figure 1 This is an optional architecture diagram of the content delivery system provided in the embodiments of this application; as shown... Figure 1 As shown, to support an information flow product application, a content delivery system 10 can be set up. In the content delivery system 10, a terminal 400 (client device, exemplarily shown as terminal 401 and terminal 402) connects to a server 200 via a network 300. The network 300 can be a wide area network (WAN), a local area network (LAN), or a combination of both. Additionally, the content delivery system 100 also includes a database 500, which provides data services to the server 200 to support content delivery by the server 200.

[0054] Database 500 can store data corresponding to each piece of content to be delivered, as well as user data corresponding to each terminal.

[0055] Terminal 401 is used to display an information flow display interface 4012 on a graphical interface 4011. The specific process of displaying the information flow display interface 4012 is as follows: Figure 2 As shown, the graphical interface 4011 of terminal 401 first displays an application list, which includes information flow product application D. When an information flow display operation is received on information flow product application D, the information flow display interface 4012 of application A is entered. During the process of entering the display interface 4012 of application A, a content retrieval request can be generated. At this time, the user corresponding to terminal 401 becomes the content requesting user, and the second feature data corresponding to the content requesting user can be obtained from terminal 401.

[0056] Terminal 402 is used to display the information flow display interface 4024 on the graphical interface 4021. The specific process of displaying the information flow display interface 4024 is as follows: Figure 3As shown, other non-information flow function interfaces, such as chat interface 4022, are first displayed on the graphical interface 4021. At the same time, an interface switching area is displayed at the bottom of the chat interface, including information flow display control 4023. When an information flow display operation is received on the information flow display control 4023, the information flow display interface 4024 is entered in response to the information flow display operation. During the process of entering the information flow display interface 4024, a content acquisition request can be generated. At this time, the user corresponding to the terminal 402 becomes the content requesting user, and then the second feature data corresponding to the content requesting user can be obtained from the terminal 402.

[0057] Server 200 is used to determine the target user volume of the target content in the next target time period based on the target exposure volume, filtering rate, and current real exposure volume at the end of the previous target time period. Simultaneously, it receives content acquisition requests initiated by various terminals 400 in the next target time period via network 300, and determines the target users associated with the target content in the next target time period based on the target user volume and target user determination strategy. Then, it filters the target users associated with the target content in the next target time period to obtain users to be delivered to, and sends the target content to these users via network 300. At the end of the next target time period, it checks whether the cumulative real exposure volume has reached the target exposure volume or whether the cumulative duration of the target time period has reached the target duration. If neither condition is met, it updates the current real exposure volume of the target content at the end of the previous target time period using the real exposure volume of the target content among the users to be delivered to, and returns to the step: Determine the target user volume of the target content in the next target time period based on the target exposure volume, filtering rate, and current real exposure volume at the end of the previous target time period. If either condition is met, the delivery process for the target content ends.

[0058] In some implementations, server 200 can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0059] Terminal 400 can be a smartphone, tablet, laptop, desktop computer, smart speaker, wearable device, smart robot, vehicle terminal, etc., but is not limited to these.

[0060] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0061] Please see Figure 4 , Figure 4 The diagram shown is a flowchart of a content delivery method according to an embodiment of this application. This method can be applied to a server (e.g., Figure 1 The method includes: (server 200 in the middle)

[0062] S110, based on the target exposure volume, filtering rate, and current actual exposure volume at the end of the previous target time period corresponding to the target content, determine the target user volume of the target content in the next target time period, where the previous target time period and the next target time period are two adjacent time periods.

[0063] As can be seen from the foregoing, there are multiple pieces of content to be delivered in the delivery system. Any one of these pieces of content can become the target content. That is, when delivery is made to a certain piece of content, that piece of content becomes the target content.

[0064] The content to be delivered can be selected by operators based on current events or rankings of top-performing content, or it can be content applied for by advertisers or content publishers. It is understood that the content delivery method in this embodiment can be a delivery method for any target content; that is, the content delivery method is applicable to any target content.

[0065] Target exposure can be understood as the amount of exposure that a certain target content actually needs to be, that is, the number of times it actually needs to appear in the client's information stream.

[0066] As one implementation method, the target exposure of the target content can be set by the operators according to the operational needs. For example, the operators can pre-set the target exposure of target content 1 to 5,000 times.

[0067] As another implementation method, some target content may carry exposure exchange information, such as virtual exchange currency, likes, collections, etc. In this case, before determining the target user volume of the target content in the next target time period based on the target exposure volume, screening rate and the current real exposure volume at the end of the previous target time period, it also includes: determining the target exposure volume of the target content based on the exposure exchange information and the conversion ratio corresponding to the exposure exchange information.

[0068] Among these, "exchange coins" can be used to redeem target exposure. Users on the information flow platform can obtain exchange coins corresponding to the target content by contributing coins. Besides exchange coins, in some cases, likes or favorites can also be used to redeem target exposure. Each type of exposure redemption information has a different conversion rate. Therefore, when target content includes exposure redemption information, the target exposure for the target content can be determined based on the exposure redemption information and its corresponding conversion rate.

[0069] For example, if the target content carries 100 exchange coins, and each exchange coin can be exchanged for 50 impressions, then the content can correspond to a total of 5,000 target impressions.

[0070] The target user base can be understood as the number of users to whom the campaign platform initially determines that the target content can be delivered; in other words, the number of users that the campaign platform needs to send to the various filtering platforms upstream. Due to the filtering rate, the target user base needs to be greater than the target exposure for the actual exposure to ultimately reach the target exposure.

[0071] The target time period refers to the time period corresponding to the content delivery method of this embodiment. In this embodiment, in order to reduce the problem of overexposure caused by latency, the number of target users can be controlled by using different target time periods. Therefore, in this embodiment, the duration of the target time period can be set in advance according to different needs, such as 1 hour, 3 hours, or 6 hours.

[0072] It is understandable that if the content delivery method of this embodiment is used for the first time to deliver a certain target content, then the next target time period after the current moment is the first target time period. Therefore, there is no previous target time period before the current moment, or it can be regarded as there is no current real exposure at the end of the previous target time period, that is, the current real exposure is 0 at the end of the previous target time period.

[0073] S120, based on the target user volume and target user determination strategy, determine the target users associated with the target content in the next target time period.

[0074] The target users associated with the target content are those initially identified by the delivery backend as the users to whom the target content can be delivered. It's understandable that in information flow product applications, the number of users requesting content received by the server in the next target time period is large, exceeding the target user count. Therefore, a target user determination strategy can be pre-set to filter the received content request users. The filtered content request users are the users associated with the target content in the next target time period. For example, users who are more suitable for or inclined to receive the target content can be selected, thereby improving the delivery effectiveness of the target content.

[0075] S130: Filter the target users associated with the target content in the next target time period to obtain the users to be targeted to, and send the target content to the users to be targeted to.

[0076] In this context, the users to be delivered to can be understood as the users to whom the target content will actually be sent.

[0077] Understandably, on the one hand, considering that the recommendation system may have already recommended too much similar content to a related target user, recommending the target content to that user again might cause resentment and hinder the expansion of content diversity for that user, thus reducing their user experience. On the other hand, considering that delivering content to a target user might affect the overall exposure data, if delivering target content to a related target user might have a significant impact on the overall exposure data, the target content will not be delivered to that related target user.

[0078] Therefore, in this embodiment, the target content is not sent to all associated target users within the next target time period. Instead, the server filters the target users associated with the target content within the next target time period to obtain the target users, and then sends the target content to each target user. For example, content can be filtered through various upstream filtering backends using weighting, scoring, sorting, and content diversity control strategies.

[0079] S140, at the end of the next target time period, use the actual exposure of the target content among the users to be targeted to update the current actual exposure of the target content at the end of the previous target time period, and return to the step: Based on the target exposure, screening rate and current actual exposure of the target content at the end of the previous target time period, determine the target user number of the target content in the next target time period, until the actual exposure at the end of the next target time period reaches the target exposure or until the cumulative duration at the end of the next target time period reaches the target duration.

[0080] The target duration can be understood as the effective duration for delivering the target content. Once the target duration is exceeded, the delivery of the target content can be stopped.

[0081] Understandably, after the above steps S110-S130, the content delivery process for the target content within the next target time period can be completed. However, for a specific target content delivery task, the actual exposure of the target content among the target users after delivery must reach the target exposure, or the delivery duration of the target content must reach the target duration. Only then is the delivery task considered complete. Therefore, if the actual exposure after the target content delivery does not reach the target exposure, or if the cumulative duration of all target time periods corresponding to the target content delivery at the end of the next target time period does not reach the target duration, the delivery of the target content needs to continue. For example, assuming the target time period is 6 hours long, and there are a total of 3 target time periods including the next target time period at the end of the next target time period, the cumulative duration corresponding to the end of the next target time period is 18 hours. If the target duration is set to 24 hours, and the actual exposure after the target content delivery does not reach the target exposure, then step S140 needs to be executed.

[0082] If the termination condition is not met at the end of the next target time period, in order to accurately determine the target user volume in the subsequent delivery of the target content, we can first use the actual exposure volume of the target content among the users to be delivered to update the current actual exposure volume of the target content at the end of the previous target time period, and then return to the step: Based on the target exposure volume, screening rate and current actual exposure volume of the target content at the end of the previous target time period, determine the target user volume of the target content in the next target time period.

[0083] It should be noted that in this embodiment, the current actual exposure at the end of the previous target time period can take many forms.

[0084] In some implementations, the current true exposure at the end of the previous target time period can be the cumulative true exposure at the end of the previous target time period, that is, the cumulative true exposure within all target time periods before the end of the previous target time period. In this case, the current true exposure at the end of the previous target time period corresponding to the target content is updated using the true exposure of the target content among the users to be targeted. This includes: using the cumulative true exposure of the target content among the users to be targeted at the end of the next target time period and the current true exposure as the updated current true exposure at the end of the previous target time period.

[0085] In this embodiment, at the end of the next target time period, the actual exposure of the target content among the users to be targeted refers to the actual exposure of the target content among the users to be targeted within the next target time period. Since the current actual exposure at the end of the previous target time period is the accumulation of the actual exposure of all target time periods before the end of the previous target time period, at the end of the next target time period, the next target time period becomes the period preceding the next target time period. At this time, when updating the current actual exposure at the end of the previous target time period, it is necessary to include the actual exposure of the adjacent previous target time period of the next target time period. That is, the accumulation of the actual exposure of the target content among the users to be targeted at the end of the next target time period and the current actual exposure is used as the updated current actual exposure at the end of the previous target time period.

[0086] For example, suppose there are four consecutive target time periods: the first target time period, the second target time period, the third target time period, and the fourth target time period. The current time is at the boundary between the second and third target time periods. At this time, the previous target time period is the second target time period, and the next target time period is the third target time period. If we want to return to the step of determining the number of target users of the target content in the next target time period based on the target exposure, filtering rate, and the current real exposure at the end of the previous target time period when the next target time period ends, then the third target time period becomes the updated previous target time period, and the fourth target time period becomes the updated next target time period. After returning to the step, the current real exposure at the end of the updated previous target time period (the third target time period) should be used. Specifically, when the third target time period ends, the cumulative real exposure of the target content among the users to be targeted during the third target time period, the real exposure of the target content among the users to be targeted during the first target time period, and the real exposure of the target content among the users to be targeted during the second target time period is used as the current real exposure at the end of the updated previous target time period (the third target time period). In other words, if you want to enter the fourth target time period and return to the previous step after the third target time period ends, the cumulative actual exposure during the first three target time periods is used as the current actual exposure at the end of the previous target time period.

[0087] Furthermore, considering that the actual client-side exposure is directly proportional to the number of target users, and that multiple pieces of content to be delivered within the same time period compete for traffic from users requesting the content, if a piece of content has overly lenient delivery parameters, more users will be able to match the delivery criteria, causing the content to reach its target exposure too quickly. This can lead to the premature termination of the delivery, resulting in a loss of competition for traffic from later users and an inability to reach higher-quality users. It can also cause uneven competition for traffic among multiple pieces of content, with intense competition in the early stages and insufficient competition in the later stages. Conversely, if a stricter scoring threshold is set, the target exposure may not be reached even after the delivery expires (i.e., the cumulative duration reaches the target duration), which will also affect the delivery effectiveness.

[0088] Therefore, to make the actual exposure more accurate and better reflect the actual competitive relationship between users requesting content, in some implementations, the current actual exposure at the end of the previous target time period can also be the actual exposure of the previous target time period at the current moment. In this case, such as Figure 5 As shown, based on the target exposure, filtering rate, and current actual exposure of the target content, the target user volume for the target content in the next target time period is determined, including the following steps:

[0089] S111, obtain historical exposure distribution data, which represents the distribution of historical exposure over time.

[0090] It is understood that in this embodiment, the historical exposure distribution data can be based on distribution data of different time lengths, such as one day, one week, or one month. The specific time length to be selected can be determined based on the timeliness of the target content. For example, for target content with strong timeliness, one day can be selected as the time length, and the corresponding historical exposure distribution data will obtain the distribution of historical exposure within one day. For target content with weak timeliness, one week or one month can be selected as the time length, and the corresponding historical exposure distribution data will obtain the distribution of historical exposure within one week or one month.

[0091] For example, a historical exposure distribution data is shown, which shows the distribution of historical exposures within a day. Assume that the total exposure of a certain day in history is 10 million, and then break it down by hour. Assume that from 0:00 to 23:00, the exposures per hour are 100,000, 20,000, 20,000, 20,000, 20,000, 50,000, 50,000, 100,000, 200,000, 300,000, 400,000, 400,000, 500,000, 500,000, 500,000, 500,000, 600,000, 800,000, 1,000,000, 1,400,000, 1,300,000, and 700,000.

[0092] S112, based on historical exposure distribution data and target exposure, determine the sub-target exposure of the target content in each target time period.

[0093] As one implementation method, after obtaining historical exposure distribution data, the proportion of exposure corresponding to each target time period within the time length can be further obtained. Then, based on the proportion of exposure corresponding to each target time period, the sub-target exposure corresponding to each target time period can be calculated when the total exposure is the target exposure.

[0094] Using the example above, if the target exposure is 10,000 and the timeliness of the target content is one day, then based on historical exposure distribution data, we can obtain the proportion of exposure for each target time period (from 0:00 to 23:00), namely 10, 2, 2, 2, 2, 2, 5, 5, 10, 20, 30, 40, 40, 50, 50, 50, 50, 50, 60, 80, 100, 140, 130, and 70. Then, combined with the target exposure, we obtain the sub-target exposure of the target content in each target time period (from 0:00 to 23:00) as 100, 20, 20, 20, 20, 20, 50, 50, 100, 200, 300, 400, 400, 500, 500, 500, 500, 500, 600, 800, 1000, 1400, 1300, and 700.

[0095] It should be noted that the above example uses one hour as the target time period. In some implementations, the target time period can be two or more hours, depending on the actual needs.

[0096] S113, based on the sub-target exposure volume within the previous target time period and the current actual exposure volume, determine the exposure volume error. The current actual exposure volume at the end of the previous target time period is the actual exposure volume within the previous target time period.

[0097] As one implementation method, the exposure of the target content can be counted in real time, thereby obtaining the actual exposure within the previous target time period, that is, the current actual exposure at the end of the previous target time period. Then, the difference between the current actual exposure at the end of the previous target time period and the sub-target exposure within the previous target time period is used as the exposure error.

[0098] Exposure error can reflect whether the target content was overexposed or underexposed in the previous target time period. Specifically, when the exposure error is greater than 0, it indicates that the target content was overexposed in the previous target time period, while when the exposure error is less than 0, it indicates that the target content was underexposed in the previous target time period.

[0099] S114. Based on the sub-target exposure, exposure error, and filtering rate of the target content in the next target time period, determine the target user volume of the target content in the next target time period.

[0100] As can be seen from the foregoing, both excessive and insufficient exposure can affect the normal competitive relationship between the target content and the users requesting the content. Furthermore, the cumulative effect over a long period is greater in the later stages of the target time period. Therefore, the number of target users in the next target time period can be adjusted at the end of the previous target time period to control the current actual exposure at the end of the next target time period, thereby reducing the impact of excessive or insufficient exposure on the normal competitive relationship between the target content and the users requesting the content.

[0101] As one implementation method, the difference between the sub-target exposure and the exposure error of the target content in the next target time period can be calculated first, and then the difference can be divided by the screening rate to obtain the target user volume of the target content in the next target time period.

[0102] For example, continuing with the above example, assuming the filtering rate is 0.5 and the current time is 1 o'clock, then the previous target time period is 0-1 o'clock, and the sub-target exposure within the previous target time period is 100. The next target time period is 1-2 o'clock, and the sub-target exposure within the next target time period is 20. And assuming the actual exposure within the previous target time period 0-1 o'clock (i.e., the current actual exposure at 1 o'clock when the previous target time period 0-1 o'clock ends) is 90, then the exposure error can be calculated to be -10, and thus the target user count of the target content in the next target time period (1-2 o'clock) can be calculated to be 60.

[0103] Accordingly, if the current true exposure at the end of the previous target time period is the same as the true exposure at the current moment in the previous target time period, the current true exposure at the end of the previous target time period corresponding to the target content is updated using the true exposure of the target content among the users to be targeted. This includes the following steps: at the end of the next target time period, the true exposure of the target content among the users to be targeted is used as the current true exposure at the end of the previous target time period corresponding to the target content.

[0104] In this embodiment, since the current actual exposure at the end of the previous target time period is the actual exposure of the previous target time period at the current moment, when the next target time period ends, the next target time period becomes the previous target time period compared to the next target time period after that. At this time, if we want to return to the step: determine the number of target users of the target content in the next target time period based on the target exposure, filtering rate and the current actual exposure at the end of the previous target time period, we can take the actual exposure of the target content among the users to be targeted in the next target time period as the current actual exposure at the end of the previous target time period adjacent to the next target time period after that.

[0105] For example, suppose there are four consecutive target time periods: the first target time period, the second target time period, the third target time period, and the fourth target time period. The current time is at the boundary between the second and third target time periods. At this time, the previous target time period is the second target time period, and the next target time period is the third target time period. If we want to return to the step of determining the number of target users of the target content in the next target time period based on the target exposure, filtering rate, and the current real exposure at the end of the previous target time period when the next target time period ends, then the third target time period becomes the updated previous target time period, and the fourth target time period becomes the updated next target time period. After returning to the step, the real exposure within the updated previous target time period (the third target time period) should be used. Specifically, at the end of the third target time period, the real exposure of the target content among the users to be targeted during the third target time period is used as the current real exposure at the end of the updated previous target time period (the third target time period). In other words, if you want to enter the fourth target time period and return to the previous step after the third target time period ends, the actual exposure during the third target time period is used as the current actual exposure at the end of the previous target time period.

[0106] The content delivery method provided in this embodiment controls the target users associated with the target content in the next target time period by using the number of target users in the next target time period. Since the target users are controlled, continuous content delivery can be avoided, reducing the total amount of content delivered and thus reducing the over-exposure. At the same time, the actual exposure is counted by the time interval of the target time period, which is equivalent to a certain waiting effect for obtaining the actual exposure, which can mitigate the impact of delay and further reduce the over-exposure.

[0107] Please see Figure 6 , Figure 6 The diagram shown is a flowchart of a content delivery method according to another embodiment of this application. This method can be applied to a server (e.g., Figure 1 The method includes: (server 200 in the middle)

[0108] S210, based on the target exposure volume, filtering rate, and current actual exposure volume at the end of the previous target time period corresponding to the target content, determine the target user volume of the target content in the next target time period, where the previous target time period and the next target time period are two adjacent time periods.

[0109] The specific implementation of step S210 can be referred to the specific description of S110, and will not be repeated here.

[0110] S220, based on the first feature information of the target content and the second feature information of the content requesting user, determine the associated user from the content requesting users.

[0111] Here, the first feature information refers to the feature information corresponding to the target content, which is information related to the target content. Optionally, the first feature information can be obtained by manually adding tags, by recognizing the target content, or by obtaining historical information corresponding to the target content.

[0112] The second feature information refers to the feature information corresponding to the content requesting user, which is content related to the content requesting user. Optionally, the second feature information can be automatically added by the content requesting user, or it can be obtained by collecting and statistically analyzing the user's historical behavior.

[0113] Associated users can be understood as users to whom the target content is suitable or inclined to be exposed. It is understood that while there are many users requesting content, the target content is not suitable or inclined to be exposed to every user. Therefore, in this embodiment, a feature information matching method can be used to initially filter out associated users from the content requesting users to obtain those to whom the target content is suitable or inclined to be exposed. Specifically, when the second feature information of a content requesting user matches the first feature information of the target content, the content requesting user can be considered an associated user to whom the target content is suitable or inclined to be exposed.

[0114] In some implementations, considering the different types of users requesting content, such as intermediate and advanced users who frequently use news feed applications, and new, casual users who occasionally use or have never used news feed applications, the content delivery method in related technology delivery systems targets the content to be delivered to users carrying corresponding tags. For example, if the tag for the content to be delivered is "sports," then the content will be delivered to users carrying the "sports" tag.

[0115] However, users carrying tags are typically mid-to-high-level users. These users don't need targeted content from a dedicated advertising system; they require recommendations from a recommendation system. Expecting too much content from a dedicated advertising system can negatively impact the recommendation system. Conversely, for new, low-level users, due to a lack of historical user behavior information, the recommendation system cannot accurately recommend content, thus requiring targeted content delivery. This leads to poor performance of existing advertising methods, failing to meet actual delivery needs. Therefore, to improve the delivery effectiveness of information flow products, some implementations, such as... Figure 7 As shown, determining associated users from content requesting users based on the first characteristic information of the target content and the second characteristic information of the content requesting users includes the following steps:

[0116] S221, retrieve the active time corresponding to the user who requested the content.

[0117] S222, Identify candidate associated users from content request users whose active time does not meet the time threshold.

[0118] S223, Based on the first feature information of the target content and the second feature information of the candidate associated users, determine the associated users from the candidate associated users.

[0119] The active time corresponding to the content requesting user can refer to the time the content requesting user uses the information feed product application. It can be the total duration or the frequency of use, such as the time spent using the information feed product application every day / week / month.

[0120] Candidate associated users can be the new shallow users mentioned above.

[0121] In this embodiment, the server can statistically record the active time corresponding to each content requesting user, and then determine whether the user is a mid-to-high-level user or a new shallow-level user based on the active time. Optionally, when the active time is the total duration, content requesting users whose active time does not meet the time threshold can be identified as candidate associated users, i.e., identified as new shallow-level users. Optionally, when the active time is the usage frequency, content requesting users whose active time does not meet the time frequency threshold can be identified as candidate associated users, i.e., identified as new shallow-level users.

[0122] Therefore, after identifying candidate related users, the associated users can be determined from the candidate related users based on the first feature information of the target content and the second feature information of the candidate related users.

[0123] In this embodiment, since the target content is selected by the operators or other content applied for for promotion, it is highly likely to be high-quality content. Therefore, the content delivery method of this embodiment can deliver more high-quality content to new and casual users, increase the proportion of high-quality content exposed to new and casual users, and thus improve the user experience of new and casual users.

[0124] S230, the previous target user quantity and associated users are used as the target users associated with the target content in the next target time period.

[0125] It is understandable that the number of associated users may be large in the next target time period after the current moment. However, in this embodiment, only the number of associated users in the previous target time period is selected as the target users associated with the target content in the next target time period.

[0126] As one implementation method, a counter can be set inside the server. The counting target of the counter is the number of target users. That is, each time an associated user is identified, that associated user is regarded as a target user, and the counter count is incremented by 1 until the counter count reaches the number of target users. At this point, after subsequent associated users are identified, they will no longer be regarded as target users. Alternatively, after receiving more content request users, the associated users will no longer be identified from the content request users based on the first feature information of the target content and the second feature information of the content request users.

[0127] S240: Filter the target users associated with the target content in the next target time period to obtain the users to be targeted, and send the target content to the users to be targeted.

[0128] S250, at the end of the next target time period, use the actual exposure of the target content among the users to be targeted to update the current actual exposure of the target content at the end of the previous target time period, and return to the step: Based on the target exposure, screening rate and the current actual exposure of the target content at the end of the previous target time period, determine the target user number of the target content in the next target time period, until the actual exposure at the end of the next target time period reaches the target exposure or until the cumulative duration at the end of the next target time period reaches the target duration.

[0129] The specific implementation of steps S240-S250 can be referred to the specific description of S130-S140, and will not be repeated here.

[0130] The content delivery method of this application embodiment determines associated users through feature information matching, which facilitates the determination of users to whom the target content is suitable or inclined to be delivered, thereby improving the accuracy of content delivery. In addition, by taking the previous target users and the associated users as the target users associated with the target content in the next target time period, the target content can be quickly sent to the users to be delivered within the target time period, thereby waiting to collect the actual exposure volume within the target time period, reducing the impact of delay, and further reducing the over-sending exposure volume.

[0131] In some implementations, the first feature information in the foregoing embodiments may include first attribute information and first historical behavior information of the user who clicked on the target content, and the second feature information may include second attribute information and second historical behavior information of the user who requested the content. In this case, please refer to [link to relevant documentation]. Figure 8 , Figure 8 The diagram shown is a flowchart of a content delivery method according to another embodiment of this application. This method can be applied to a server (e.g., Figure 1 The method includes: (server 200 in the middle)

[0132] S310, based on the target exposure volume, filtering rate, and current actual exposure volume at the end of the previous target time period corresponding to the target content, determine the target user volume of the target content in the next target time period, where the previous target time period and the next target time period are two adjacent time periods.

[0133] The specific implementation of step S310 can be referred to the specific description of S110, and will not be repeated here.

[0134] S320, if the second attribute information matches the first attribute information, determine the content requesting user as a candidate user.

[0135] Based on the foregoing, the first feature information includes first attribute information and the first historical behavior information of the user who clicked on the target content. The first attribute information can also be understood as the tag information corresponding to the target content. For example, the first attribute information may include the age group / gender / category / tags / interest profile corresponding to the target content, where the age group / gender / category / tags / interest profile corresponding to the target content can be interpreted as the age group / gender / category / tags / interest profile of users to whom the target content is suitable or inclined to be exposed.

[0136] The primary attribute information of the target content can be obtained in multiple ways.

[0137] Optionally, the first attribute information can be determined manually. For example, the operators or promoters who select the content to be delivered can manually determine and add the corresponding first attribute information to the content based on the title or body of the content to be delivered. Thus, when a certain content to be delivered becomes the target content, the target content can carry the first attribute information.

[0138] Optionally, the first attribute information can also be predicted by a neural network model. In this case, such as Figure 9 As shown, the first attribute information can be obtained through the following steps:

[0139] S321, Obtain descriptive information corresponding to the target content.

[0140] S322, the neural network model is used to predict the attributes of descriptive information to obtain the first attribute information output by the neural network model. The neural network model is trained by descriptive information carrying attribute information labels.

[0141] The descriptive information corresponding to the target content can be the title or body text of the target content. The body text can include text, images, or videos.

[0142] After obtaining the descriptive information corresponding to the target content, the descriptive information can be input into the neural network model, and the neural network model can then predict the attributes of the descriptive information to obtain the first attribute information output by the neural network model.

[0143] In the process of training a neural network model, descriptive information carrying attribute information labels can be input into the initial neural network model. The initial neural network model can output predicted attribute information based on the descriptive information and calculate the loss between the predicted attribute information and the attribute information labels of the descriptive information. The model parameters of the initial neural network model are adjusted according to the loss, thereby obtaining the trained neural network model.

[0144] The second attribute information can also be understood as the tag information of the content requesting user. For example, it may include the age group / gender / category / tags / interest profile of the content requesting user. Among them, the age group / gender / category / tags / interest profile of the content requesting user can be actively filled in by the content requesting user when registering the information feed account, or it can be collected and statistically obtained by the content requesting user during the use of the information feed application.

[0145] Candidate users can be understood as users to whom the target content is suitable or inclined to be exposed. Since the first attribute information can be interpreted as information about users to whom the target content is suitable or inclined to be exposed, and the second attribute information can be understood as the target user's tag information, if the second attribute information matches the first attribute information, it indicates that the target content is suitable or inclined to be exposed to content request users with the second attribute information, i.e., candidate users.

[0146] For example, suppose the first attribute information of a target content is <18-20 years old, male, sports, ...>, and there are two content requesting users, namely user A and user B. User A's second attribute information is <19 years old, male, sports>, and user B's second attribute information is <19 years old, female, entertainment>. In this case, it is determined that user A's second attribute information matches the first attribute information of the target content, thereby identifying user A as a candidate user.

[0147] S330, extract features from the second historical behavior information of the candidate users to obtain the second feature vector corresponding to the candidate users, and extract features from the first historical behavior information of the users who clicked on the target content to obtain the first feature vector.

[0148] The second historical behavior information of candidate users may include historical information corresponding to the candidate user's exposure and click behavior.

[0149] The first historical behavior information of users who click on the target content can include the exposure of the users who clicked on the target content and the historical information corresponding to the click behavior.

[0150] As one implementation method, feature extraction algorithms can be used to extract features.

[0151] Optionally, the feature extraction algorithm can be LDA (Linear Discriminant Analysis). Specifically, the second historical behavior information of candidate users can be used as input parameters to obtain a 256-dimensional feature vector, i.e., the second feature vector, output by the LDA algorithm. The first historical behavior information of each user who clicked on the target content can be used as input parameters to obtain 256-dimensional sub-feature vectors corresponding to each user who clicked on the target content. Then, these sub-feature vectors are pooled and merged to obtain the first feature vector.

[0152] Understandably, some content to be advertised may have already been clicked or exposed before being selected. For example, operators might choose content with high click-through rates for further promotion to increase its exposure, or promoters might choose content already recommended by the system but with low exposure to further increase its reach. In such cases, the content has already been clicked, and therefore, the server can store the first historical behavior information of users who clicked on that content. Thus, when the content is identified as the target content, the server can directly obtain the first characteristic information, including the first attribute information, and the first historical behavior information of the users who clicked on the target content.

[0153] In other cases, the content to be delivered may be cold-start content, which refers to content without historical exposure or click data. Therefore, for some content to be delivered, it may not have historical exposure or click data before being selected as the target content. For example, newly generated content. Thus, when the content to be delivered is determined as the target content, it does not carry the first historical behavior information of the users who clicked the target content. In this case, in order to extract features from the first historical behavior information of the users who clicked the target content and obtain the first feature vector, it is necessary to first obtain the first historical behavior information of the users who clicked the target content. In some implementations, such as... Figure 10 As shown, Figure 10 The flowchart illustrates the process of obtaining the first historical behavior information of users who clicked on the target content. Specifically, before feature extraction from the first historical behavior information of users who clicked on the target content, the following steps may also be included:

[0154] S331, Obtain the second attribute information that matches the first attribute information of the target content.

[0155] S332, obtain the target number of historical users with second attribute information as users who clicked on the target content.

[0156] It is understandable that the server can obtain a large amount of historical user data, for example, from a database connected to the server. Considering that historical users with second attribute information matching the first attribute information of the target content are highly likely to click on the target content after it is exposed, in this embodiment, for cold-start target content where no users have clicked beforehand, in the initial stage, historical users with second attribute information matching the first attribute information of the cold-start target content can be selected, thus treating these users as users who clicked on the cold-start target content. Therefore, for cold-start target content, the subsequent step of feature extraction of the first historical behavior information of users who clicked on the target content can also be performed. Here, historical users refer to users who have historically used the information flow product application; these users can carry second attribute information.

[0157] S340, Based on the similarity between the first feature vector and the second feature vector and the similarity threshold, determine the associated users from the candidate users.

[0158] Understandably, even though the number of users requesting content has been limited to some extent through attribute information matching in the latter target time period, the number of remaining candidate users is still large. Therefore, the number of candidate users can be further limited to identify related users from among them.

[0159] In this embodiment, similarity can be used as a limiting factor, along with a pre-set similarity threshold. Specifically, the similarity between the first and second feature vectors is calculated, and then compared to the pre-set similarity threshold. If the similarity between the second feature vector of a candidate user and the first feature vector of the target content is greater than the similarity threshold, the candidate user is identified as an associated user, thus allowing the identification of associated users from among the candidate users. Optionally, the similarity threshold can be set based on experience.

[0160] Optionally, the similarity between the first feature vector and the second feature vector can be the cosine distance. After obtaining the cosine distance, the cosine distance is normalized to obtain a similarity score, which ranges from [0,1]. The larger the similarity score, the more similar the first feature vector and the second feature vector are.

[0161] Furthermore, it is understood that the decision to send a particular piece of content to a user is made only upon receiving the user's content request, thus identifying that content as the target content. Therefore, when the server receives a content request from a user, it can determine the target content based on that user. In some implementations, the target content is one of a predetermined number of candidate content items ranked highest in similarity among multiple candidate content items corresponding to the associated user. These candidate content items are those whose similarity to the content to be delivered meets a similarity threshold.

[0162] In this embodiment, when a content requesting user is obtained, the system first searches for content to be delivered that matches the second attribute information of the content requesting user from all the content to be delivered. Then, it calculates the similarity between the first feature vector of these corresponding content to be delivered and the second feature vector of the content requesting user. The content to be delivered with a similarity greater than a similarity threshold is determined as the candidate content corresponding to the content requesting user. Then, a preset number of candidate content with the highest similarity ranking are selected from the candidate content as the target content associated with the content requesting user. At this time, the content requesting user also becomes the associated user of the target content. That is, whenever a content requesting user is obtained, the target content associated with the content requesting user can be determined. The target content is one of the preset number of candidate content with the highest similarity ranking among the multiple candidate content corresponding to the associated user. The candidate content is the content to be delivered whose similarity meets the similarity threshold among the content to be delivered.

[0163] S350, the number of associated users of the previous target user group, is used as the target user group for the target content in the next target time period.

[0164] The specific implementation of step S350 can be referred to the specific description of S230, and will not be repeated here.

[0165] S360 filters the target users associated with the target content in the next target time period to obtain the users to be targeted, and sends the target content to the users to be targeted.

[0166] S370, at the end of the next target time period, use the actual exposure of the target content among the users to be targeted to update the current actual exposure of the target content at the end of the previous target time period, and return to the step: Based on the target exposure, screening rate and current actual exposure of the target content at the end of the previous target time period, determine the target user number of the target content in the next target time period, until the cumulative actual exposure at the end of the next target time period reaches the target exposure or until the cumulative duration at the end of the next target time period reaches the target duration.

[0167] The specific implementation of steps S360-S370 can be found in the detailed description of S130-S140, and will not be repeated here.

[0168] Furthermore, on the one hand, considering that the same content delivery task ID may exist at different time points, but these identical task IDs correspond to different tasks, this leads to potential competition between identical task IDs. Therefore, in actual tasks, it is desirable to complete the delivery of the target content within the previous target time period. If the delivery of the target content is not completed within the previous target time period, i.e., the actual exposure volume does not reach the target exposure volume, it is also desirable to complete the delivery of the target content within the subsequent target time period. On the other hand, as mentioned above, in order to make the actual exposure volume more accurate and more in line with the actual competitive relationship between users requesting content, in some implementations, the target user volume of the target content in the subsequent target time period can be determined based on the sub-target exposure volume, exposure volume error, and filtering rate of the target content in the subsequent target time period. In this case, the target user volume in the subsequent target time period may be adjusted based on the sub-target user volume in the subsequent target time period. Therefore, it is desirable that the actual exposure volume in the subsequent target time period can adapt to the adjusted target user volume.

[0169] Therefore, regardless of the considerations mentioned above, the speed of exposure generation can be altered by adjusting the similarity threshold. It's understandable that the smaller the similarity threshold, the more users request content that meets the threshold, resulting in a faster increase in exposure. That is, in some implementations, before the step of determining the target user count for the target content in the next target time period based on the target exposure, filtering rate, and current actual exposure, the step of updating the similarity threshold at the end of the next target time period is also included.

[0170] Optionally, if the goal is to complete the delivery of the target content within the next target time period, the similarity threshold can be updated at the end of the next target time period. Specifically, the similarity threshold can be reduced at the end of the next target time period, so that more users request content that meets the similarity threshold, thereby increasing the exposure faster, so as to quickly replenish the remaining exposure required for the delivery task in the next target time period.

[0171] Optionally, if it is desired that the actual exposure volume of the next target time period, adjacent to the previous target time period, is adapted to the adjusted target user volume at the end of the next target time period, then updating the similarity threshold at the end of the next target time period can be specifically determined by the magnitude of the exposure volume error of the target content at the end of the next target time period. Specifically, if the exposure volume error is greater than 0, the similarity threshold is increased; if the exposure volume error is less than 0, the similarity threshold is decreased.

[0172] Furthermore, it is understandable that after the target content is sent to the target users, some of the target users may click on the target content. In this case, in some implementations, after sending the target content to the target users, the steps may also be included: recording the users who clicked on the target content among the target users, and updating the list of users who clicked on the target content based on the users who clicked on the target content.

[0173] After the target content is sent to the users to be targeted, the target content may be exposed on the users' terminal devices. At this time, some users may be interested in the target content and click on it. After a user clicks on the target content, the server can record the user who clicked on the target content. The server can then update the user information of the user who clicked on the target content and retrieve the updated historical behavior information of the user who clicked on the target content from the database. Thus, even for target content that has been cold-started, the first historical behavior information of the user who accurately clicked on the target content can be obtained in the content delivery method of this embodiment.

[0174] The content delivery method provided in this application calculates the feature similarity between the second historical behavior information of candidate users and the first historical behavior information of users who click on target content, and sets a similarity threshold to filter content request users. On the one hand, it can filter out more suitable users to send to, further improving the accuracy of content delivery. On the other hand, it can control the speed at which target content is sent to users to be delivered, avoiding sending target content to users too quickly, increasing the proportion of high-quality users, or quickly adapting to the real exposure required for the next target time period.

[0175] The following will describe an exemplary application of the embodiments of this application in a real-world application scenario.

[0176] This invention is applied to the "highlights" feature of social media software, establishing a delivery system to deliver content that requires a smaller exposure.

[0177] Typically, the delivery system stores a certain number of content items to be delivered. In this embodiment, it is assumed to be 1000 (including content items X, Y, Z, etc.). Each content item corresponds to task information such as target exposure, target time period, and target duration, as well as first attribute information (such as age group / gender / category / tags / interest profile, etc.) and the first historical behavior information of the user who clicked on the content item.

[0178] Next, the delivery system can monitor and obtain content requests initiated by users in real time. For example, after obtaining a content request user A, it can further obtain the active time corresponding to the content request user A, as well as the corresponding secondary attribute information (such as age group / gender / category / tags / interest profile, etc.) and secondary historical behavior information.

[0179] Next, the delivery system can compare the active time of the content requesting user A with a time threshold. If it is less than the time threshold, the content requesting user A is determined to be a candidate associated user. Then, the first attribute information of 1,000 content items to be delivered is matched with the second attribute information of the content requesting user A to obtain the matching content items to be delivered. For example, there are 50 matching content items to be delivered.

[0180] Then, for each of the 50 content items to be delivered, the first historical behavior information of the user who clicked on the content item is input into the LDA algorithm to obtain the first feature vector for each content item. The second historical behavior information of the user A who requested the content is also input into the LDA algorithm to obtain the second feature vector for user A. The similarity between the first feature vector of each content item and the second feature vector of user A is calculated, and users with similarity scores greater than a similarity threshold (e.g., 0.75) are selected. For example, if 15 users remain, then the top N content items with the highest similarity scores can be selected from these 15, for example, the top 3. These 3 content items can then become the target content.

[0181] The steps above demonstrate how the delivery system determines the target content when it receives a content request from user A. The process of determining the target content after the delivery system receives other content request users can be referenced above.

[0182] After determining the target content (e.g., target content X, Y, Z) corresponding to content request user A, since content request user A has matched target content X, Y, Z, the counters for target content X, Y, and Z can be incremented by 1 respectively, and target content X, Y, and Z can be sent to the filtering backend. The filtering backend will then filter target content X, Y, and Z through weighting, scoring, sorting, and content diversity control strategies, thereby sending the filtered target content (e.g., target content X and Y) to content request user A.

[0183] Since the delivery processes for target content X, Y, and Z are similar, the following explanation will focus on the delivery process for target content X.

[0184] The above process is repeated continuously. After a period of time, such as 2 hours (less than the target time period corresponding to the target content, such as 6 hours), assuming that the counter of the target content X reaches the counting target, that is, the target number of users, such as 2000, when it is the first time to deliver the target content X, the current real exposure of the previous target time period is 0. Then the delivery system can calculate the target number of users of the target content based on the target exposure of the target content X, the filtering rate, and the current real exposure of the previous target time period (assuming the target exposure of the target content X is 1000 and the filtering rate is 0.5, then the target number of users = (1000-0) / 0.5 = 2000). That is, the target content X needs to become the target content corresponding to 2000 content request users through the above steps.

[0185] Once the counter for target content X reaches its target, processing of target content X is paused. This means that after a user requests content again, content X will no longer be considered the target content for that user. This continues until the target time period, i.e., 6 hours, at which point the actual exposure of target content X within those 6 hours is calculated, let's say 800.

[0186] Since the current actual exposure of target content X within 6 hours is 800, which is less than the target exposure of 1000, additional exposure is needed for target content X. At this point, the counter for target content X needs to be reset to reflect the target number of users. The specific calculation is: Target number of users = (1000 - 800) / 0.5 = 400. Furthermore, since the goal of subsequently supplementing the exposure of target content X is to reach the target number of users (400) as quickly as possible, the similarity threshold can be reduced (e.g., from 0.75 to 0.7).

[0187] After a period of time, such as 0.5 hours (less than the target time period corresponding to the target content, such as 6 hours), assuming the counter for target content X reaches the counting target, i.e., the target number of users, such as 400, meaning target content X needs to become the target content for 400 users requesting content through the aforementioned steps. Until the target time period has elapsed again, i.e., another 6 hours have elapsed, the current actual exposure of target content X in the previous two 6-hour periods is calculated, assuming it reaches 1000, after which the delivery process for target content X can be terminated.

[0188] Furthermore, if the above process is repeated until the target duration (e.g., 24 hours, or 4 target time periods) is reached, and the cumulative current real exposure of target content X within the 4 target time periods still does not reach 1000, then the delivery process for target content X will also end.

[0189] The following section will further illustrate the content delivery method provided in this application embodiment by taking a user request as an example, i.e., when the server receives a content request from a user. For example... Figure 11 As shown, the content delivery method of this application includes the following steps:

[0190] S410, acquire data.

[0191] The data includes task data such as the target exposure, target time period, and target duration of the content to be delivered, first attribute information and first historical behavior information of users who clicked on the content to be delivered, second attribute information and second historical behavior information of users who requested the content, and active duration of users who requested the content.

[0192] S420: Determine whether the user requesting the content is a new, inexperienced user. If so, proceed to step S430. Otherwise, end the process for that user.

[0193] New, shallow-level users can be understood as candidate related users. Users who request content and whose active duration is less than a time threshold are identified as new, shallow-level users.

[0194] S430, Feature Similarity Score.

[0195] The first historical behavior information and the second historical behavior information are characterized, and the similarity scores are calculated between the features corresponding to the first historical behavior information and the features corresponding to the second historical behavior information.

[0196] S440: Does the similarity score exceed the similarity threshold? If it does, proceed to step S450; otherwise, end the process.

[0197] S450: Obtain the top N content items with the highest similarity scores and use them as the target content for the user requesting the content.

[0198] S460, determine whether the content requesting user has exceeded the upper limit of the counter corresponding to the target content. If yes, proceed to step S470; otherwise, proceed to step S480.

[0199] The upper limit of the counter count can be understood as the number of target users corresponding to the target content.

[0200] S470: Determine if the target time period has ended. If yes, proceed to step S490; otherwise, end the process.

[0201] S480, increment the counter corresponding to the target content by 1, and then end.

[0202] S490, the counter corresponding to the target content is updated with its upper limit and count value, and the process ends.

[0203] It should be noted that the above-mentioned examples of specific implementable methods provided in this application can be arbitrarily combined without conflict to form a new content delivery method. It should be understood that any new content delivery method formed by combining any of the examples should fall within the protection scope of this application.

[0204] It should also be noted that in some alternative implementations, the execution order of some processes in the embodiments of the present invention may differ from the execution order described in the foregoing specific embodiments. For example, two consecutive processes may actually be executed in parallel, or they may sometimes be executed in reverse order, depending on the functions involved.

[0205] Please see Figure 12 , Figure 12 A block diagram of a content delivery device 500 according to an embodiment of this application is shown. The device 500 includes: a target user volume determination module 510, a target user determination module 520, a filtering and sending module 530, and a loop execution module 540.

[0206] The target user volume determination module 510 is used to determine the target user volume of the target content in the next target time period based on the target exposure volume, filtering rate and the current real exposure volume at the end of the previous target time period. The previous target time period and the next target time period are two adjacent time periods.

[0207] The target user identification module 520 is used to identify the target users associated with the target content in the next target time period based on the target user volume and the target user identification strategy.

[0208] The filtering and sending module 530 is used to filter the target users associated with the target content in the next target time period, obtain the users to be delivered to, and send the target content to the users to be delivered to.

[0209] The loop execution module 540 is used to update the current real exposure of the target content at the end of the previous target time period by using the actual exposure of the target content among the users to be targeted at the end of the next target time period, and return the step: Based on the target exposure, screening rate and current real exposure of the target content at the end of the previous target time period, determine the target user number of the target content in the next target time period, until the cumulative real exposure at the end of the next target time period reaches the target exposure or until the cumulative duration at the end of the next target time period reaches the target duration.

[0210] In one implementation, the target user determination module 520 includes: an associated user determination submodule and a target user determination submodule.

[0211] The associated user determination submodule is used to determine associated users from content request users based on the first feature information of the target content and the second feature information of the content request user.

[0212] The target user determination submodule is used to identify the previous target user group and associated users as the target users associated with the target content in the next target time period.

[0213] In one implementation, the first feature information includes first attribute information and first historical behavior information of the user who clicked the target content; the second feature information includes second attribute information and second historical behavior information of the content requesting user; the associated user determination submodule is further configured to determine the content requesting user as a candidate user when the second attribute information matches the first attribute information; extract features from the second historical behavior information of the candidate user to obtain a second feature vector corresponding to the candidate user, and extract features from the first historical behavior information of the user who clicked the target content to obtain a first feature vector; and determine associated users from the candidate users based on the similarity between the first feature vector and the second feature vector and a similarity threshold.

[0214] As one implementation, the loop execution module 540 is also used to update the similarity threshold at the end of the next target time period.

[0215] As one implementation method, the target content is one of a preset number of candidate contents ranked first in similarity among multiple candidate contents corresponding to the associated user, and the candidate content is the content to be delivered whose similarity meets the similarity threshold.

[0216] As one implementation method, the target content is cold start content, and the associated user determination submodule is also used to obtain second attribute information that matches the first attribute information of the target content; and to obtain a target number of historical users with the second attribute information as users who clicked on the target content.

[0217] As one implementation method, the associated user determination submodule is also used to record users who clicked on the target content among the users to be targeted, and to update the users who clicked on the target content using the users who clicked on the target content.

[0218] As one implementation method, the first attribute information is obtained through the following steps: obtaining descriptive information corresponding to the target content; performing attribute prediction on the descriptive information through a neural network model to obtain the first attribute information output by the neural network model, wherein the neural network model is trained by the descriptive information carrying attribute information labels.

[0219] As one implementation, the associated user determination submodule is also used to obtain the active time corresponding to the content requesting user; determine candidate associated users from the content requesting users whose active time does not meet the time threshold; and determine the associated user from the candidate associated users based on the first feature information of the target content and the second feature information of the candidate associated users.

[0220] As one implementation method, the current actual exposure at the end of the previous target time period is the accumulation of the actual exposure at the end of the previous target time period. The loop execution module 540 is also used to take the accumulation of the actual exposure of the target content among the users to be delivered at the end of the next target time period and the current actual exposure as the updated current actual exposure at the end of the previous target time period.

[0221] In one implementation, the target user volume determination module 510 is further configured to acquire historical exposure distribution data, which characterizes the distribution of historical exposure over time; based on the historical exposure distribution data and the target exposure, determine the sub-target exposure of the target content within each target time period; based on the sub-target exposure in the previous target time period and the current actual exposure, determine the exposure error, with the current actual exposure at the end of the previous target time period being the actual exposure within that period; and based on the sub-target exposure, exposure error, and filtering rate of the target content in the next target time period, determine the target user volume of the target content in the next target time period. Correspondingly, the loop execution module 540 is further configured to, at the end of the next target time period, use the actual exposure of the target content among the users to be targeted as the current actual exposure at the end of the previous target time period corresponding to the target content.

[0222] In one implementation, the device 500 further includes a target exposure determination module, used to determine the target exposure corresponding to the target content based on exposure exchange information and the conversion ratio corresponding to the exposure exchange information.

[0223] The content delivery device provided in this application embodiment controls the target users associated with the target content in the next target time period by using the number of target users in the next target time period. Since the target users are controlled, continuous content delivery can be avoided, the total amount of content delivered can be reduced, and thus the over-exposure can be reduced. At the same time, the actual exposure is counted by the time interval of the target time period, which is equivalent to a certain waiting effect for obtaining the actual exposure, which can mitigate the impact of delay and further reduce the over-exposure.

[0224] It should be noted that the device embodiments in this application correspond to the aforementioned method embodiments. The specific principles in the device embodiments can be found in the content of the aforementioned method embodiments, and will not be repeated here.

[0225] The following will combine Figure 13 This application describes one type of server.

[0226] Please see Figure 13 Based on the above content delivery method, this application embodiment also provides another server 100 including a processor 102 capable of executing the aforementioned method.

[0227] The server 100 also includes a memory 104. The memory 104 stores programs that can execute the contents of the foregoing embodiments, and the processor 102 can execute the programs stored in the memory 104.

[0228] The processor 102 may include one or more cores for data processing and message matrix units. The processor 102 connects to various parts of the server 100 via various interfaces and lines, and performs various functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 104, and by calling data stored in the memory 104. Optionally, the processor 102 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 102 may integrate one or more of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the displayed content; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 102 and may be implemented separately using a communication chip.

[0229] The memory 104 may include random access memory (RAM) or read-only memory (ROM). The memory 104 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 104 may include a program storage area and a data storage area. The program storage area may store instructions for implementing an operating system, instructions for implementing at least one function, instructions for implementing the various method embodiments described below, etc. The data storage area may also store data acquired by the server 100 during use (e.g., data to be recommended and operating methods).

[0230] Server 100 may also include a network module and a screen. The network module is used to receive and transmit electromagnetic waves, converting electromagnetic waves into electrical signals, thereby enabling communication with communication networks or other devices, such as audio playback devices. The network module may include various existing circuit elements used to perform these functions, such as antennas, radio frequency transceivers, digital signal processors, encryption / decryption chips, user identity module (SIM) cards, memory, etc. The network module can communicate with various networks such as the Internet, corporate intranets, and wireless networks, or communicate with other devices via wireless networks. The aforementioned wireless networks may include cellular telephone networks, wireless local area networks, or metropolitan area networks. The screen can display interface content and facilitate data interaction.

[0231] In some embodiments, the server 100 may further include a peripheral interface and at least one peripheral device. The processor 102, memory 104, and peripheral interface 106 can be connected via a bus or signal line. Each peripheral device can be connected to the peripheral interface via a bus, signal line, or circuit board. Specifically, the peripheral device includes at least one of the following: a radio frequency component 108, a positioning component 112, a camera 114, an audio component 116, a display screen 118, and a power supply 122.

[0232] Peripheral interface 106 can be used to connect at least one I / O (Input / Output) related peripheral device to processor 102 and memory 104. In some embodiments, processor 102, memory 104 and peripheral interface 106 are integrated on the same chip or circuit board; in some other embodiments, any one or two of processor 102, memory 104 and peripheral interface 106 can be implemented on separate chips or circuit boards, and this application embodiment does not limit this.

[0233] The radio frequency (RF) component 108 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The RF component 108 communicates with communication networks and other communication devices via electromagnetic signals. The RF component 108 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals back into electrical signals. Optionally, the RF component 108 includes: an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a user identity module card, etc. The RF component 108 can communicate with other terminals via at least one wireless communication protocol. This wireless communication protocol includes, but is not limited to: the World Wide Web, metropolitan area networks, intranets, various generations of mobile communication networks (2G, 3G, 4G, and 5G), wireless local area networks, and / or WiFi (Wireless Fidelity) networks. In some embodiments, the RF component 108 may also include circuitry related to NFC (Near Field Communication), which is not limited in this application.

[0234] The positioning component 112 is used to locate the current geographic location of the server in order to enable navigation or LBS (Location Based Service). The positioning component 112 can be a positioning component based on GPS (Global Positioning System), BeiDou system or Galileo system.

[0235] Camera 114 is used to capture images or videos. Optionally, camera 114 includes a front-facing camera and a rear-facing camera. Typically, the front-facing camera is located on the front panel of server 100, and the rear-facing camera is located on the back of server 100. In some embodiments, there are at least two rear-facing cameras, which are any one of a main camera, a depth-sensing camera, a wide-angle camera, and a telephoto camera, to achieve background blurring by fusion of the main camera and the depth-sensing camera, panoramic shooting by fusion of the main camera and the wide-angle camera, VR (Virtual Reality) shooting, or other fusion shooting functions. In some embodiments, camera 114 may also include a flash. The flash can be a single-color temperature flash or a dual-color temperature flash. A dual-color temperature flash refers to a combination of a warm light flash and a cool light flash, which can be used for light compensation at different color temperatures.

[0236] Audio component 116 may include a microphone and a speaker. The microphone is used to collect sound waves from the user and the environment, and convert the sound waves into electrical signals that are input to processor 102 for processing, or input to radio frequency component 108 for voice communication. For stereo acquisition or noise reduction purposes, there may be multiple microphones, each located at a different part of server 100. The microphone may also be an array microphone or an omnidirectional microphone. The speaker is used to convert electrical signals from processor 102 or radio frequency component 108 into sound waves. The speaker may be a conventional diaphragm speaker or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, it can convert electrical signals not only into sound waves that humans can hear, but also into sound waves that humans cannot hear for purposes such as ranging. In some embodiments, audio component 114 may also include a headphone jack.

[0237] Display screen 118 is used to display a UI (User Interface). This UI may include graphics, text, icons, videos, and any combination thereof. When display screen 118 is a touch display screen, it also has the ability to collect touch signals on or above its surface. These touch signals can be input as control signals to processor 102 for processing. In this case, display screen 118 can also be used to provide virtual buttons and / or a virtual keyboard, also known as soft buttons and / or a soft keyboard. In some embodiments, there may be one display screen 118, which serves as the front panel of server 100; in other embodiments, there may be at least two display screens, respectively disposed on different surfaces of server 100 or in a folded design; in still other embodiments, display screen 118 may be a flexible display screen, disposed on a curved or folded surface of server 100. Furthermore, display screen 118 may be configured as a non-rectangular, irregular shape, i.e., a non-rectangular screen. Display screen 118 may be made of materials such as LCD (Liquid Crystal Display) or OLED (Organic Light-Emitting Diode).

[0238] Power supply 122 is used to power the various components in server 100. Power supply 122 can be AC ​​power, DC power, a disposable battery, or a rechargeable battery. When power supply 122 includes a rechargeable battery, the rechargeable battery can be a wired rechargeable battery or a wireless rechargeable battery. A wired rechargeable battery is a battery that is charged via a wired line, while a wireless rechargeable battery is a battery that is charged via a wireless coil. The rechargeable battery can also be used to support fast charging technology.

[0239] This application also provides a computer-readable storage medium. This computer-readable medium stores program code that can be called by a processor to execute the methods described in the above method embodiments.

[0240] Computer-readable storage media can be electronic storage devices such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM. Optionally, computer-readable storage media includes non-transitory computer-readable storage medium. The computer-readable storage medium has storage space for program code that performs any of the method steps described above. This program code can be read from or written to one or more computer program products. The program code can be compressed, for example, in a suitable form.

[0241] This application also provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods described in the various optional implementations above.

[0242] In summary, the content delivery method, apparatus, server, storage medium, and computer program product or computer program provided in this application control the target users associated with the target content within the next target time period by controlling the number of target users within that time period. Because the target users are controlled, continuous content delivery can be avoided, reducing the total amount of content delivered and thus reducing over-exposure. At the same time, the actual exposure is counted based on the time interval of the target time period, which is equivalent to a certain waiting effect in obtaining the actual exposure, which can mitigate the impact of delay and further reduce over-exposure.

[0243] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A content delivery method, characterized in that, include: Based on the target exposure volume, screening rate, and current actual exposure volume at the end of the previous target time period corresponding to the target content, the target user volume of the target content in the next target time period is determined, wherein the previous target time period and the next target time period are two adjacent time periods; the screening rate refers to the proportion of users used to filter out the content for delivery. Based on the first feature information of the target content and the second feature information of the content requesting user, determine the associated user from the content requesting user; The number of previously target users and the associated users are used as the target users associated with the target content in the next target time period. The target users associated with the target content in the next target time period are filtered to obtain the users to be targeted, and the target content is sent to the users to be targeted; At the end of the next target time period, the current real exposure of the target content at the end of the previous target time period is updated using the actual exposure of the target content among the users to be targeted, and the process returns to the step: Based on the target exposure, screening rate and current real exposure of the target content at the end of the previous target time period, the target number of the target content in the next target time period is determined until the actual exposure at the end of the next target time period reaches the target exposure or until the cumulative duration at the end of the next target time period reaches the target duration. The step of determining associated users from content requesting users based on the first feature information of the target content and the second feature information of the content requesting users includes: If the second attribute information in the second feature information matches the first attribute information in the first feature information, the content requesting user is determined to be a candidate user. The first feature information also includes the first historical behavior information of the user who clicked the target content, and the second feature information also includes the second historical behavior information of the content requesting user. Feature extraction is performed on the second historical behavior information of the candidate user to obtain a second feature vector corresponding to the candidate user, and feature extraction is performed on the first historical behavior information of the user who clicked the target content to obtain a first feature vector. Based on the similarity between the first feature vector and the second feature vector and a similarity threshold, the associated user is determined from the candidate users. or, The step of determining associated users from content requesting users based on the first feature information of the target content and the second feature information of the content requesting users includes: obtaining the active time corresponding to the content requesting users, wherein the active time refers to the total time the content requesting users use the information flow product application within a set period; determining candidate associated users from content requesting users whose active time does not meet the time threshold; and determining associated users from the candidate associated users based on the first feature information of the target content and the second feature information of the candidate associated users.

2. The method according to claim 1, characterized in that, Before determining the target user count for the target content in the next target time period based on the target exposure, filtering rate, and current actual exposure of the target content in the return step, the method further includes: At the end of the next target time period, the similarity threshold is updated.

3. The method according to claim 1, characterized in that, The target content is one of a predetermined number of candidate contents ranked first in similarity among multiple candidate contents corresponding to the associated user, and the candidate content is the content to be delivered whose similarity meets the similarity threshold.

4. The method according to claim 1, characterized in that, The target content is cold-start content. Before extracting features from the first historical behavior information of the user who clicked the target content, the method further includes: Obtain second attribute information that matches the first attribute information of the target content; The target number of historical users with the second attribute information are obtained as the users who clicked the target content.

5. The method according to claim 4, characterized in that, After sending the target content to the users to be targeted, the method further includes: Record the users among the users to be targeted who clicked on the target content, and update the list of users who clicked on the target content based on the users who clicked on the target content.

6. The method according to claim 1, characterized in that, The first attribute information is obtained through the following steps: Obtain descriptive information corresponding to the target content; The descriptive information is used to predict attributes through a neural network model to obtain the first attribute information output by the neural network model. The neural network model is trained using descriptive information carrying attribute information labels.

7. The method according to any one of claims 1-6, characterized in that, The current actual exposure at the end of the previous target time period is the cumulative actual exposure at the end of the previous target time period. Updating the current actual exposure at the end of the previous target time period corresponding to the target content using the actual exposure of the target content among the users to be targeted includes: The sum of the actual exposure of the target content among the users to be targeted at the end of the next target time period and the current actual exposure is used as the current actual exposure at the end of the previous target time period after the update.

8. The method according to any one of claims 1-6, characterized in that, The determination of the target user volume for the target content in the next target time period based on the target exposure volume, filtering rate, and current actual exposure volume corresponding to the target content includes: Obtain historical exposure distribution data, which characterizes the distribution of historical exposure over time. Based on the historical exposure distribution data and the target exposure, the sub-target exposure of the target content in each target time period is determined; Based on the sub-target exposure volume within the previous target time period and the current actual exposure volume, the exposure volume error is determined, and the current actual exposure volume at the end of the previous target time period is the actual exposure volume within the previous target time period. Based on the sub-target exposure, exposure error, and filtering rate of the target content in the next target time period, the target user volume of the target content in the next target time period is determined. The step of updating the current real exposure of the target content at the end of the previous target time period by utilizing the actual exposure of the target content among the users to be targeted includes: At the end of the next target time period, the actual exposure of the target content among the users to be targeted is taken as the current actual exposure of the target content at the end of the previous target time period.

9. The method according to claim 1, characterized in that, The target content carries exposure redemption information. Before determining the target user volume of the target content in the next target time period based on the target exposure volume, filtering rate, and current actual exposure volume at the end of the previous target time period, the method further includes: Based on the exposure conversion information and the corresponding conversion ratio, the target exposure amount corresponding to the target content is determined.

10. A content delivery device, characterized in that, include: The target user volume determination module is used to determine the target user volume of the target content in the next target time period based on the target exposure volume, the filtering rate, and the current actual exposure volume at the end of the previous target time period. The previous target time period and the next target time period are two adjacent time periods. The filtering rate refers to the proportion of users selected for delivery of the content. The associated user determination submodule is used to determine associated users from content request users based on the first feature information of the target content and the second feature information of the content request user; The target user determination submodule is used to identify the previous target user and the associated users as the target users associated with the target content in the next target time period. The filtering and sending module is used to filter the target users associated with the target content in the next target time period, obtain the users to be delivered to, and send the target content to the users to be delivered to. The loop execution module is used to update the current real exposure of the target content at the end of the previous target time period by using the actual exposure of the target content among the users to be targeted at the end of the next target time period, and return to the step: Based on the target exposure, screening rate and current real exposure of the target content at the end of the previous target time period, determine the target user number of the target content in the next target time period, until the cumulative real exposure at the end of the next target time period reaches the target exposure or until the cumulative duration corresponding to the end of the next target time period reaches the target duration; In the step of determining associated users from content requesting users based on the first feature information of the target content and the second feature information of the content requesting user, the associated user determination submodule is configured to: determine the content requesting user as a candidate user if the second attribute information in the second feature information matches the first attribute information in the first feature information; the first feature information further includes the first historical behavior information of the user who clicked the target content, and the second feature information further includes the second historical behavior information of the content requesting user; perform feature extraction on the second historical behavior information of the candidate user to obtain a second feature vector corresponding to the candidate user, and perform feature extraction on the first historical behavior information of the user who clicked the target content to obtain a first feature vector; and determine the associated user from the candidate users based on the similarity between the first feature vector and the second feature vector and a similarity threshold. or, The associated user determination submodule is used to: obtain the active time corresponding to the content requesting user, wherein the active time refers to the total time that the content requesting user uses the information flow product application within a set period; Candidate associated users are identified from content request users whose active time does not meet the time threshold; associated users are identified from the candidate associated users based on the first feature information of the target content and the second feature information of the candidate associated users.

11. The apparatus according to claim 10, characterized in that, The loop execution module is further configured to update the similarity threshold at the end of the next target time period.

12. The apparatus according to claim 10, characterized in that, The target content is one of a predetermined number of candidate contents ranked first in similarity among multiple candidate contents corresponding to the associated user, and the candidate content is the content to be delivered whose similarity meets the similarity threshold.

13. The apparatus according to claim 10, characterized in that, The target content is cold start content, and the associated user determination submodule is further used to: obtain second attribute information that matches the first attribute information of the target content; The target number of historical users with the second attribute information are obtained as the users who clicked the target content.

14. The apparatus according to claim 13, characterized in that, The associated user determination submodule is further configured to: record users among the users to be targeted who clicked on the target content, and update the users who clicked on the target content using the users who clicked on the target content.

15. The apparatus according to claim 10, characterized in that, The first attribute information is obtained through the following steps: Obtain descriptive information corresponding to the target content; The descriptive information is used to predict attributes through a neural network model to obtain the first attribute information output by the neural network model. The neural network model is trained using descriptive information carrying attribute information labels.

16. The apparatus according to any one of claims 10-15, characterized in that, The current actual exposure at the end of the previous target time period is the cumulative actual exposure at the end of the previous target time period. The loop execution module is further used to: take the cumulative actual exposure of the target content among the users to be targeted at the end of the next target time period and the current actual exposure as the updated current actual exposure at the end of the previous target time period.

17. The apparatus according to any one of claims 10-15, characterized in that, The target user volume determination module is also used for: Obtain historical exposure distribution data, which characterizes the distribution of historical exposure over time. Based on the historical exposure distribution data and the target exposure, the sub-target exposure of the target content in each target time period is determined; Based on the sub-target exposure volume within the previous target time period and the current actual exposure volume, the exposure volume error is determined, and the current actual exposure volume at the end of the previous target time period is the actual exposure volume within the previous target time period. Based on the sub-target exposure, exposure error, and filtering rate of the target content in the next target time period, the target user volume of the target content in the next target time period is determined. Correspondingly, the loop execution module is also used to: at the end of the next target time period, take the actual exposure of the target content among the users to be targeted as the current actual exposure of the target content at the end of the previous target time period.

18. The apparatus according to claim 10, characterized in that, The content delivery device further includes a target exposure determination module, which is used to determine the target exposure corresponding to the target content based on the exposure exchange information and the conversion ratio corresponding to the exposure exchange information.

19. A server, characterized in that, include: One or more processors; Memory; One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs being configured to perform the method as described in any one of claims 1-9.

20. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains program code that can be invoked by a processor to execute the method as described in any one of claims 1-9.

21. A computer program product, characterized in that, It includes computer instructions, which, when executed by a processor, implement the method as described in any one of claims 1-9.

Images