Content delivery method, apparatus, electronic device and storage medium

Abstract

The application relates to the computer technical field, and particularly relates to a content delivery method and device, electronic equipment and storage medium, to improve the accuracy of the delivery object screening and improve the content delivery efficiency. The method comprises the following steps: obtaining a set of to-be-delivered contents; obtaining the content representation features corresponding to each to-be-delivered content based on the content information of each to-be-delivered content; obtaining the object representation features of the seed objects associated with each to-be-delivered content; determining the target representation features corresponding to each to-be-delivered content based on the content representation features and the object representation features corresponding to each to-be-delivered content; obtaining the target objects corresponding to each to-be-delivered content based on the target representation features, and delivering each to-be-delivered content to the target objects corresponding to each to-be-delivered content. The application jointly recalls the target objects from the content and behavior dimensions, which are used for subsequent content delivery, and can effectively improve the accuracy of the delivery object screening and the content delivery efficiency.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a content delivery method, apparatus, electronic device, and storage medium. Background Technology

[0002] With the development of internet technology, more and more content creation platforms have emerged. In the current online environment, the user can be either a content viewer or a content creator. When the user acts as a content creator, they can publish content online through content creation platforms, making the content presented online increasingly personalized and diversified.

[0003] Taking text and image content as an example, text and image content includes images and the text corresponding to those images. After creators publish text and image content through content creation platforms, the advertising team needs to ensure that the text and image content can obtain the corresponding exposure through advertising.

[0004] In related technologies, the target audience and target volume for each piece of content to be delivered in the task list are generally set, and the most relevant target audience (traffic) for the task is recalled as the target audience.

[0005] However, this strategy is mostly used for precise targeting or cold start warm-up. Therefore, when determining the target audience, it relies more on prior knowledge, such as the existing fans of the content to be targeted and the tags associated with the task. As a result, the selected target audience may not actually view the content, thus affecting the final effect of the campaign. Summary of the Invention

[0006] This application provides a content delivery method, apparatus, electronic device, and storage medium to improve the accuracy of content delivery target selection and increase content delivery efficiency.

[0007] This application provides a content delivery method, including:

[0008] Obtain a set of content to be delivered, wherein the set of content to be delivered includes at least one piece of content to be delivered;

[0009] Based on the content information of each content to be delivered, obtain the corresponding content representation features; and obtain the object representation features of at least one seed object associated with each content to be delivered; the seed object is a historical object whose sorting result is in the first order range, determined based on the historical behavior of the object in the initial delivery process of the corresponding content to be delivered.

[0010] Based on the content representation features corresponding to each content to be delivered, and at least one object representation feature, the target representation features corresponding to each content to be delivered are determined respectively;

[0011] Based on the representation features of each target, obtain multiple target objects corresponding to each content to be delivered, and deliver each content to its corresponding target object.

[0012] This application provides a content delivery device, comprising:

[0013] A content acquisition unit is used to acquire a set of content to be delivered, wherein the set of content to be delivered includes at least one piece of content to be delivered.

[0014] The first representation unit is used to obtain the corresponding content representation features based on the content information of each content to be delivered; and to obtain the object representation features of at least one seed object associated with each content to be delivered; wherein the seed object is a historical object whose sorting result is within the first order range, determined based on the historical behavior of the object in the initial delivery process of the corresponding content to be delivered.

[0015] The second representation unit is used to determine the target representation features corresponding to each content to be delivered based on the content representation features corresponding to each content to be delivered, and at least one object representation feature.

[0016] The content delivery unit is used to obtain multiple target objects corresponding to each content to be delivered based on each target representation feature, and to deliver each content to be delivered to its corresponding target object.

[0017] Optionally, the content delivery unit is specifically used for:

[0018] For each piece of content to be delivered, perform the following operations:

[0019] Obtain a first set of candidate objects associated with the target representation features of a content to be delivered, and a set of historical objects associated with the content to be delivered, the set of historical objects including at least one historical object that has viewed the content to be delivered, and the object representation features of the candidate objects in the first set of candidate objects are ranked in the second order range with the correlation between the target representation features and the object representation features.

[0020] Based on the historical object set, the first candidate object set is filtered to obtain the second candidate object set;

[0021] The candidate objects in the second candidate object set are divided into layers, and the candidate objects with the target layer are taken as the target objects.

[0022] Optionally, the content delivery unit:

[0023] Obtain the behavior information of each candidate object in the second candidate object set within a preset historical time period;

[0024] Based on the behavioral information of the candidate objects, the candidate objects are stratified.

[0025] Optionally, the behavioral information of each candidate object within a preset historical time period shall include at least:

[0026] The number of visits within the preset historical time period;

[0027] Total content browsing time within the preset historical time period;

[0028] Click-through rate within the preset historical time period.

[0029] Optionally, the content delivery unit is further configured to:

[0030] If the ratio of the number of candidate objects with target level to the target delivery volume corresponding to the content to be delivered is less than the target value, then a new first candidate object set associated with the target representation features of the content to be delivered is obtained again, and candidate objects with target level are determined again from the new first candidate object set as target objects, until the ratio of the total number of determined target objects to the target delivery volume is not less than the target value.

[0031] Wherein, the target value is less than the ratio of the number of candidate objects in the first candidate object set to the target delivery quantity.

[0032] Optionally, the second representation unit is specifically used for:

[0033] Based on at least one object representation feature corresponding to each of the content to be delivered, determine the behavioral representation feature corresponding to each of the content to be delivered;

[0034] Based on the content representation features and behavior representation features corresponding to each of the content to be delivered, the target representation features corresponding to each of the content to be delivered are determined.

[0035] Optionally, the second representation unit is specifically used for:

[0036] Linear interpolation is performed on the content representation features and behavior representation features corresponding to each content to be delivered to obtain the target representation features corresponding to each content to be delivered.

[0037] Optionally, the content delivery unit is specifically used for:

[0038] Based on the multiple target objects corresponding to each of the content to be delivered, a sequence of content to be delivered corresponding to each target object is determined.

[0039] Based on the delivery weight of each content in the sequence of content to be delivered, content is delivered to the target object, whereby the delivery weight represents the degree of correlation between the corresponding content to be delivered and the target object.

[0040] An electronic device provided in this application includes a processor and a memory, wherein the memory stores program code, and when the program code is executed by the processor, the processor performs the steps of any of the above-described content delivery methods.

[0041] This application provides a computer-readable storage medium including program code. When the program code is run on an electronic device, the program code is used to cause the electronic device to perform the steps of any of the above-described content delivery methods.

[0042] This application provides a computer program product or a computer program. The computer program product includes computer instructions stored in a computer-readable storage medium. When the processor of an electronic device reads the computer instructions from the computer-readable storage medium, the processor executes the computer instructions, causing the electronic device to perform the steps of any of the above-described content delivery methods.

[0043] The beneficial effects of this application are as follows:

[0044] This application provides a content delivery method, apparatus, electronic device, and storage medium. In determining the delivery target for the content to be delivered, this application combines two dimensions: content and behavior. Specifically, from a content perspective, it obtains the content representation features of each content to be delivered. From a behavior perspective, based on relevant information during the initial delivery process, it determines the object representation features of at least one associated seed object for each content. Then, it combines the content and behavior dimensions, combining the content and object representation features, to determine the target representation features corresponding to each content. Finally, it filters and selects matching target objects as delivery targets based on these target representation features. This approach, based on multi-dimensional information for user recall—specifically, jointly recalling target objects from both content (prior knowledge) and behavior (posterior knowledge) dimensions for subsequent content delivery—effectively improves the accuracy of target object selection and increases content delivery efficiency.

[0045] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0046] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0047] Figure 1 This is a schematic diagram of a content delivery method in the related technology of this application;

[0048] Figure 2 This is an optional schematic diagram of an application scenario in an embodiment of this application;

[0049] Figure 3 This is a flowchart illustrating the implementation of a content delivery method in an embodiment of this application.

[0050] Figure 4 This is a schematic diagram of a matrix decomposition method according to an embodiment of this application;

[0051] Figure 5 This is a schematic diagram of a method for calculating a target vector in an embodiment of this application;

[0052] Figure 6 This is a flowchart illustrating an implementation method for representing content to be delivered in an embodiment of this application.

[0053] Figure 7 This is a flowchart illustrating the implementation of a target object screening method in an embodiment of this application.

[0054] Figure 8 This is a flowchart illustrating the implementation of a text and image delivery method according to an embodiment of this application.

[0055] Figure 9 This is a schematic diagram of a number packet construction process in an embodiment of this application;

[0056] Figure 10 This is a schematic diagram of another number packet construction process in an embodiment of this application;

[0057] Figure 11 This is a schematic diagram of the composition structure of a content delivery device according to an embodiment of this application;

[0058] Figure 12 This is a schematic diagram of the hardware structure of an electronic device using an embodiment of this application;

[0059] Figure 13 This is a schematic diagram of the hardware structure of another electronic device using an embodiment of this application. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this application. Obviously, the described embodiments are only some embodiments of the technical solutions of this application, and not all embodiments. Based on the embodiments recorded in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the technical solutions of this application.

[0061] The following describes some of the concepts involved in the embodiments of this application.

[0062] Content to be distributed: This refers to content that content creators publish on a content creation platform, which can then be distributed through a distribution team to gain sufficient exposure. The content in this application can be text and image content, such as news articles, or videos, short videos, etc.

[0063] Content information: refers to the information contained in the content to be delivered. For example, the content information of text and image content refers to the images and text contained in the content, while the content information of video content refers to the video cover, video frames, etc.

[0064] Representation Features: This application proposes an embedding feature, which can be understood as the representation feature of each entity in space, and can be in vector form. Specifically, this application includes: content representation features, object representation features, behavior representation features, and target representation features. Among them, the content representation feature is the vector representation of the content to be delivered, also known as the content vector; the object representation feature is the vector representation of the candidate object and the target object, also known as the user vector; the behavior representation feature refers to the behavior vector obtained by performing certain calculations on the aforementioned user vectors; and the target representation feature refers to the target vector obtained by performing certain calculations on the aforementioned behavior vector and content vector.

[0065] Matrix decomposition (factorization) involves breaking down a matrix into a product of several matrices. It can be categorized into triangular factorization, full-rank factorization, QR factorization, Jordan factorization, and Singular Value Decomposition (SVD), with the three most common being triangular factorization, QR factorization, and singular value decomposition. In this embodiment, user vectors can be obtained through matrix decomposition.

[0066] Target reach: The amount of exposure (traffic) each piece of content to be delivered needs to obtain from the content creation platform. In this embodiment, each piece of content to be delivered x * All are associated with a target delivery volume y * This means that the task needs to give z * Exposure to users on various platforms. Related terms to target delivery volume include: guaranteed delivery, over-delivery, and under-delivery. Guaranteed delivery means ensuring the target delivery volume for the content to be delivered is achieved; over-delivery means the exposure volume of the task exceeds the target delivery volume by more than 20%; under-delivery means the exposure volume of the task is less than the target delivery volume by more than 20%.

[0067] Number Packet: A collection of user IDs used for real-time delivery, consisting of the target user IDs of the content to be delivered.

[0068] The technical solutions in this application involve artificial intelligence (AI) and machine learning (ML) technologies, and are designed based on computer vision technology and machine learning in artificial intelligence.

[0069] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence.

[0070] Artificial intelligence (AI) studies the design principles and implementation methods of various intelligent machines, enabling them to perceive, reason, and make decisions. AI technology mainly includes computer vision, natural language processing, machine learning / deep learning, autonomous driving, and intelligent transportation. With the research and advancement of AI technology, it is being researched and applied in multiple fields, such as smart homes, intelligent customer service, virtual assistants, smart speakers, intelligent marketing, autonomous driving, robotics, and smart healthcare. It is believed that with further technological development, AI will be applied in even more fields and play an increasingly important role.

[0071] Machine learning is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory, among others. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Compared to data mining, which focuses on finding patterns in large datasets, machine learning emphasizes algorithm design, enabling computers to automatically "learn" patterns from data and use these patterns to predict unknown data.

[0072] Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence. Its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, and inductive learning. The model in the embodiments of this application was trained using machine learning or deep learning techniques.

[0073] The content delivery method proposed in this application mainly consists of two parts: model training and model application. The model training part involves the field of machine learning technology, which is used to train a recall model and update its parameters. After the model training is completed, the trained recall model can be obtained using the above method. Then, based on the recall model, a set of candidate objects related to each task to be delivered can be obtained. Furthermore, user filtering and hierarchical user selection are performed on the candidate object set to obtain the final target users for delivery.

[0074] The design concept of the embodiments of this application is briefly introduced below:

[0075] With the development of internet technology, more and more content creation platforms have emerged. Creators can publish content online through these platforms, making the content presented online increasingly personalized and diverse. After creators publish multimedia content through these platforms, the distribution team needs to ensure that the multimedia content achieves the corresponding exposure through distribution.

[0076] In related technologies, to ensure sufficient delivery volume, the delivery strategy listed in the background technology is generally adopted, which can be simply referred to as the offline number package strategy. Specifically, the processing flow of the offline number package strategy is as follows: Figure 1 As shown:

[0077] The system retrieves the list of tasks to be deployed offline for the day. Based on the target audience and target volume set for each piece of content in the task list, it recalls several target users most relevant to the task and generates a corresponding number of phone number packages. Finally, it generates a corresponding offline phone number package for each piece of content to be deployed, which will be used for real-time deployment to achieve the deployment goals of the current task.

[0078] Because this strategy requires offline calculation of the phone number packages before the task goes live, it cannot accurately guarantee the required delivery volume for each piece of content. Therefore, the offline calculation process involves a significant amount of redundant computation. Furthermore, most offline phone number package delivery strategies are used for targeted advertising or cold start warm-up. Thus, when acquiring user phone number packages, offline methods rely heavily on prior knowledge, such as the existing followers of the account corresponding to the content to be delivered and the tags associated with the task. They lack subsequent data, such as users' actual consumption behavior regarding the content to be delivered, which affects the final delivery results.

[0079] In view of this, embodiments of this application propose a content delivery method, apparatus, electronic device, and storage medium. In determining the delivery target corresponding to the content to be delivered in this embodiment, both content and behavioral dimensions are combined. Specifically, from the content dimension, the content representation features of each content to be delivered are obtained. Furthermore, from the behavioral dimension, based on relevant information during the initial delivery process of each content to be delivered, the object representation features of at least one associated seed object are determined. Then, the content and behavioral dimensions are combined, and the content and object representation features are combined to determine the target representation features corresponding to each content to be delivered. Finally, the matching target objects are selected as delivery targets based on these target representation features. This approach, by jointly recalling target objects from both the content dimension (prior knowledge) and the behavioral dimension (posterior knowledge) for subsequent content delivery, can effectively improve the accuracy of delivery target selection and increase content delivery efficiency.

[0080] The preferred embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit this application. Furthermore, the embodiments and features in the embodiments of this application can be combined with each other without conflict.

[0081] like Figure 2The diagram shown is an application scenario illustration of an embodiment of this application. The application scenario diagram includes two terminal devices 210 and one server 220.

[0082] In this embodiment, the terminal device 210 includes, but is not limited to, mobile phones, tablets, laptops, desktop computers, e-book readers, smart voice interaction devices, smart home appliances, and in-vehicle terminals. The terminal device may have a client installed related to image and text delivery. This client can be software (e.g., a browser), a webpage, or a mini-program. The server 220 is the backend server corresponding to the software, webpage, or mini-program, or a server specifically used for image and text delivery; this application does not impose specific limitations. The server 220 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing 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.

[0083] It should be noted that the content delivery method in this embodiment can be executed by an electronic device, which can be a server 220 or a terminal device 210. That is, the method can be executed by the server 220 or the terminal device 210 alone, or by both the server 220 and the terminal device 210. When the terminal device 210 executes the method alone, for example, the terminal device 210 obtains the content representation features of each content to be delivered in the set of content to be delivered, as well as the object representation features of the related seed objects. Then, based on these features, it determines the target representation features corresponding to each content to be delivered, and filters the target objects based on the target representation features. When the server 220 executes the method alone, it does so in a similar manner. The server 220 obtains the content representation features of each content to be delivered in the set of content to be delivered, as well as the object representation features of the related seed objects. Then, based on these features, it determines the target representation features corresponding to each content to be delivered, and filters the target objects based on the target representation features. When the terminal device 210 and the server 220 execute together, one optional implementation is that the terminal device 210 obtains the content representation features of each content to be delivered in the set of content to be delivered, as well as the object representation features of the related seed objects. Then, based on these features, it determines the target representation features corresponding to each content to be delivered and sends them to the server 220. The server 220 then filters target objects based on the target representation features. In the following explanation, the example mainly uses the server 220 executing alone, and no specific limitations are made.

[0084] It should be noted that the offline number package generation strategy for content to be delivered proposed in this application embodiment jointly recalls the target users corresponding to the delivery task through the content dimension of the content to be delivered and the historical user behavior dimension of the task, for use in subsequent content delivery scenarios.

[0085] In this embodiment, the content delivery scenario can specifically include news, trending topics, short videos, etc. This paper primarily uses the image / text coin-giving application scenario as an example. A typical image / text coin-giving application scenario is as follows: users give coins to their favorite image / text content; correspondingly, the content creator can convert the received coins into corresponding traffic based on the platform's exchange rate, and hand it over to the delivery team for subsequent content delivery support. On some content delivery platforms, the delivery team can further promote the content based on the converted traffic. Similarly, the delivery team needs to ensure that the content receiving coins gains corresponding exposure through delivery, and minimize over- or under-delivery.

[0086] In one alternative implementation, the terminal device 210 and the server 220 can communicate via a communication network.

[0087] In one alternative implementation, the communication network is a wired network or a wireless network.

[0088] It should be noted that, Figure 2 The examples shown are merely illustrative; in reality, the number of terminal devices and servers is unlimited and is not specifically limited in the embodiments of this application.

[0089] In this embodiment of the application, when there are multiple servers, the multiple servers can form a blockchain, and the servers are nodes on the blockchain; as disclosed in the content delivery method of this embodiment, the content information, feature data, etc. involved can be stored on the blockchain, such as the content representation features of each content to be delivered, the object representation features of related seed objects, etc.

[0090] Furthermore, the embodiments of this application can be applied to various scenarios, including but not limited to cloud technology, artificial intelligence, smart transportation, and assisted driving.

[0091] The following describes the content delivery method provided by the exemplary embodiments of this application in conjunction with the application scenarios described above and with reference to the accompanying drawings. It should be noted that the above application scenarios are only shown to facilitate understanding of the spirit and principles of this application, and the embodiments of this application are not limited in any way in this respect.

[0092] See Figure 3 The diagram shown is a flowchart of an implementation method for content delivery provided in this application. The specific implementation process of this method is as follows:

[0093] S31: The server obtains a set of content to be delivered, which includes at least one piece of content to be delivered;

[0094] The content to be published can be text and images, such as news and articles, or videos, such as short videos. This article mainly uses text and images as an example for illustration.

[0095] For example, given a list of content to be delivered: l = [x1, x2, ..., x k ], where L is a set of content to be delivered, containing k content items, also known as delivery tasks, and k represents the number of image and text tasks to be delivered that day, and each content item x * All are associated with a target delivery volume z * This means that the task needs to give z * Expose users on various platforms.

[0096] After obtaining the set of content to be delivered, the target audience (also known as target users) can be identified based on the target delivery volume of each content to be delivered, so as to carry out subsequent content delivery. The specific process is shown in steps S32 and S33.

[0097] S32: The server obtains the corresponding content representation features of each content to be delivered based on the content information of each content to be delivered; and obtains the object representation features of at least one seed object associated with each content to be delivered.

[0098] Seed objects are historical objects whose sorting results fall within the first order range, determined based on their historical behavior during the initial delivery of the corresponding content to be delivered.

[0099] It should be noted that for a given piece of content to be delivered, it has already received some exposure on the platform before the target user recall is performed based on steps S32 and S33 (e.g., users consume the content and generate coins). Therefore, before large-scale exposure, some users have consumed this content, and these users can be referred to as historical objects, or historical users. In this embodiment, some users can be selected as seed objects based on the historical behavior of these historical objects.

[0100] In the embodiments of this application, the first order range can be determined according to actual needs. In this article, the sorting result is taken as the first 100 as an example.

[0101] Content to be released x *For example, based on the exposure that the task has gained on the platform in the early stages, the historical behavior information of the corresponding users can be obtained. For example, the consumption time of historical users who have viewed the task can be obtained. Then, based on the consumption time, the users are sorted and the top 100 users are selected as seed objects. The top 100 users can be considered as being in the first order range.

[0102] It should be noted that the various representation features involved in the embodiments of this application can be in the form of vectors.

[0103] For example, the content representation features of the content to be delivered can be represented by the content vector x. * The object representation feature of the seed object can be represented by the user vector u. * It means that an x * There are 100 corresponding u * Therefore, for each piece of content to be delivered, it can be based on the corresponding x * And 100 u * This involves determining the target representation features corresponding to each piece of content to be delivered, also referred to below as the target vector, denoted as...

[0104] The following example uses text and image content in the form of an article to illustrate this:

[0105] Specifically, in the target user recall phase, it can be implemented based on a recall model. In order to obtain the vector representation of each piece of content to be delivered, the Word2vec method is used in this embodiment to obtain the word vector e of each word in the title and body of the article. * The learning process involves using TF-IDF (term frequency–inverse document frequency) combined with Word2vec to learn the embedding representation of each article, i.e., the content vector. For article x... k Content vector x k The calculation formula is as follows:

[0106]

[0107] Where, x k This indicates the content to be delivered (article x). k The content vector representation of article x, where i ranges from 0 to m. k The i-th word, m is the total number of words, tf k,i and IDF i They represent word i in article x. k In terms of word frequency and semantic weight, e iThis represents the word vector of word i learned using Word2vec.

[0108] After obtaining article x k The vector representation of x k Then, by utilizing users' consumption behavior of text and image content on the platform over a past period (e.g., three days), matrix factorization can be used to derive the user vector representation u for each corresponding seed object using a fixed article vector. * .

[0109] See Figure 4 As shown, it is a schematic diagram of a matrix decomposition method in an embodiment of this application.

[0110] In image and text delivery scenarios, an item is an article, and the article content includes tags, primary categories, secondary categories, body text, and titles. Among these, the body text is the most content-rich. This application uses the article body text as a corpus, and after word segmentation, it learns the word vector for each word using Word2vec. For each article added to the corpus, this application can obtain its word2vec vector using a weighted word vector method. At a certain moment, this application's article corpus contains M articles, and the vectors of these M articles constitute an article matrix, denoted as:

[0111] Y = [y1, ..., y M ];

[0112] Among them, y j Let be the vector of the j-th article. Let the historical behavior data of user u be:

[0113] P u =[r u1 ,…,r uM ];

[0114] Where, r uj This indicates whether user u clicked on article j, with 0 indicating no click and 1 indicating a click. The historical behavioral data of different users u for different articles can be used to construct this data. Figure 4 The rating matrix R shown is shown.

[0115] In this embodiment of the application, the vector x of user u can be obtained by solving the following incremental matrix factorization model. u :

[0116]

[0117] Among them, c uj =1+αr uj α and λ are hyperparameters.

[0118] It's important to emphasize that the article matrix here was pre-calculated using word2vec and is treated as a constant in the objective function; only x needs to be solved. u Based on the above formula, the vector x of user u can be obtained. u That is, the user vector u in this paper * .

[0119] After obtaining the content (article) vector x * and user vector u * Subsequently, as mentioned above, this application jointly recalls users similar to the content to be delivered as target users based on both the content dimension of the content to be delivered and the behavior dimension of historical users, as detailed below.

[0120] S33: The server determines the target representation features corresponding to each content to be delivered based on the content representation features corresponding to each content to be delivered, as well as at least one object representation feature;

[0121] Among them, the target representation feature is the target vector corresponding to the content to be delivered.

[0122] See Figure 5 The diagram illustrates a method for calculating target vectors according to an embodiment of this application. Assuming each piece of content to be delivered corresponds to 100 user vectors, taking text and image content 1 as an example, the calculation can be based on the 100 user vectors u corresponding to that content. j First, determine the behavior vector b1 corresponding to the content. Then, based on the content vector x1 and the behavior vector b1, the target vector corresponding to the text and image content 1 can be calculated. Using a similar method, the target vector for each piece of content to be delivered can be obtained. The process is described in detail below:

[0123] One alternative implementation is to proceed as follows: Figure 6 The flowchart shown implements step S33. (See attached document.) Figure 6 The diagram shown is an implementation flowchart of a content representation method in this application embodiment, which specifically includes the following sub-steps:

[0124] S601: The server determines the behavioral representation features corresponding to each content to be delivered based on at least one object representation feature corresponding to each content to be delivered.

[0125] Optionally, step S601 can be implemented by averaging, that is:

[0126] The behavioral representation features of each content to be delivered are obtained by averaging the object representation features corresponding to each content to be delivered.

[0127] For example, given the content to be delivered x k It is possible to obtain its content vector representation x. k Furthermore, since the coin-tossing task has accumulated certain user behaviors (coin tossing, reading, etc.) on the platform before the task is issued, in this embodiment, for each piece of content to be distributed, the top 100 users by consumption time are collected as seed objects. Then, by analyzing the current content to be distributed x... k The average of the user vectors of the corresponding seed objects is used to obtain the behavior vector representation b of the content to be delivered. k :

[0128]

[0129] Where j takes values ​​from 1 to 100, u j Indicates the content to be delivered (x). k The corresponding user vector of the j-th seed object, b k That is, the content to be delivered x k The corresponding behavioral characteristics.

[0130] S602: The server determines the target representation features corresponding to each content to be delivered based on the content representation features and behavior representation features corresponding to each content to be delivered.

[0131] Optionally, step S602 can be implemented using linear interpolation, i.e.:

[0132] Linear interpolation is performed on the content representation features and behavior representation features corresponding to each content to be delivered to obtain the target representation features corresponding to each content to be delivered.

[0133] In this embodiment of the application, after obtaining the content vector x of the content to be delivered... k and behavior vector b k Then, linear interpolation is used to obtain the content x to be delivered. k Final vector representation:

[0134]

[0135] Here, α and β represent the weights of the content vector and the behavior vector, respectively.

[0136] In this embodiment of the application, the values ​​of α and β can both be 0.5, and the vector... Content to be delivered (x) k The final vector representation, i.e., the content x to be delivered. k The target vector.

[0137] It should be noted that the values ​​of α and β listed above are only illustrative examples. In actual applications, they can be flexibly set according to the actual situation, and no specific restrictions are made here.

[0138] In the above implementation, when recalling target users for each piece of content to be delivered, not only is the content information of the image and text task (title, body text, and tags, etc.) considered, but also the historical behavioral information of the image and text task is effectively incorporated by using linear interpolation, taking into account the characteristics of the coin-toss task. This can better recall target users related to the content to be delivered, build offline number packages, and further improve the delivery performance indicators of the content to be delivered, such as the click-through rate of the image and text.

[0139] S34: The server obtains multiple target objects corresponding to each content to be delivered based on the representation features of each target, and delivers each content to be delivered to its corresponding target object.

[0140] Compared to offline number package strategies in related technologies, this application's embodiment combines two dimensions—content and behavior—when determining the target audience for the content to be delivered. Specifically, from a content perspective, it obtains the content representation features of each content item; from a behavior perspective, it determines the object representation features of at least one associated seed object based on relevant information from the initial delivery process of each content item. Then, it combines the content and behavior dimensions, using both content and object representation features, to determine the target representation features corresponding to each content item. Finally, it filters and selects matching target objects as delivery targets based on these target representation features. This approach, by jointly recalling target objects from both content (prior knowledge) and behavior (posterior knowledge) dimensions for subsequent content delivery, effectively improves the accuracy of target selection and increases content delivery efficiency.

[0141] The process of obtaining the target object in the embodiments of this application will be described in detail below:

[0142] The acquisition process can be specifically divided into several modules: user recall, user filtering, user segmentation, and tiered user screening. The following will combine these modules... Figure 7 The flowchart illustrating the target object screening method provides a detailed explanation of the process:

[0143] One alternative implementation is to proceed as follows: Figure 7 The flowchart shown illustrates the process of obtaining multiple target objects corresponding to each piece of content to be delivered in step S34. For each piece of content to be delivered, the following steps are executed:

[0144] S701: The server obtains a first set of candidate objects associated with the target representation information of a content to be delivered, and a set of historical objects associated with the content to be delivered;

[0145] The historical object set includes at least one historical object that has viewed the content to be delivered, and the correlation between the object representation information and the target representation information of the candidate objects in the first candidate object set is ranked within the second order range.

[0146] Specifically, for the content x to be delivered * The user recall module can recall the associated first candidate object set, wherein the obtained first candidate object set The candidate objects in the dataset are the sum vectors recalled based on cosine similarity. The users corresponding to the top n most similar user vectors constitute the first candidate set, which is the target user candidate set. Within this set, the phone number packets associated with the content to be delivered are an ordered list. The higher a user's ranking within a phone number packet, the stronger the correlation between the user and the current content to be delivered, as shown below:

[0147]

[0148] in, Indicates the content to be delivered (x). * The corresponding first candidate object sets, u1 and w *,1 Representing task x respectively * The most relevant target user u and the weight w associated with that user, w is the targeting weight in this paper.

[0149] In this embodiment, w can be represented as the cosine similarity between the target representation feature of the content to be delivered and the object representation feature of the object, that is, the distance between the target vector of the content to be delivered and the user vector. The closer the distance between the vectors, the higher the similarity and the larger the corresponding w value.

[0150] In this embodiment, considering that the user recall module in step S701 may recall users who have already consumed the content to be delivered as candidate users, in order to prevent the content to be delivered from being exposed to platform users a second time, this application solves this problem by setting up a user filtering module. The specific implementation of this module is described in step S702:

[0151] S702: The server filters the first candidate object set based on the historical object set to obtain the second candidate object set;

[0152] Specifically, for all content to be delivered, a list of historical consumers can be retrieved. That is, the set of historical objects in this article, all of which belong to the historical consumers related to the corresponding content to be delivered;

[0153] One possible implementation is to, for each piece of content x to be delivered... * By filtering out the first candidate object set In the list of historical consumers User u in the set can obtain the second candidate object set. In the specific implementation process, step S702 can be further divided into the following sub-steps:

[0154] S7021: The server compares the historical object set with the first candidate object set;

[0155] S7022: Based on the comparison results, the server deletes the candidate objects located in the historical object set from the first candidate object set and obtains the second candidate object set.

[0156] For example, given the content to be delivered x k The corresponding first candidate object set is:

[0157]

[0158] The corresponding set of historical objects is:

[0159]

[0160] lie in The candidate objects are u1 and u2, therefore based on right After filtering, the second set of candidate objects is as follows:

[0161]

[0162] In the above implementation, after passing through the user filtering module, a preliminary pre-offline number package can be obtained for each piece of content to be delivered.

[0163] It should be noted that the above-listed... This is just a simple example. In actual implementation, to ensure the target quantity is met and to prevent the number of target users remaining after filtering from being less than the target delivery volume, the number of users N in the first candidate set that can be recalled is four times the target delivery volume of the content to be delivered. For example, for the content to be delivered x... k The corresponding target delivery volume z k When the number is 100,000, the number of users in the first candidate object set is N = 400,000.

[0164] Furthermore, unlike the offline number package strategy in related technologies, which only relies on prior knowledge of the content to be delivered to recall users, this application uses both content and behavioral dimensions to jointly recall the target users, obtaining... Furthermore, by setting up a user filtering module, the platform users are prevented from being exposed to the campaign again, thus obtaining an initial pre-offline number package.

[0165] It should be noted that, depending on the business characteristics, this application may employ different features or models for recall when combining content and behavioral dimensions. However, regardless of which features or models are used for recall, it falls under the category of user recall based on multi-dimensional information, and this application does not impose specific limitations here.

[0166] Furthermore, in order to better achieve the goal of maintaining user volume and avoid negative impacts on the platform's overall metrics, this application proposes a user segmentation sub-strategy. When generating offline number packages for users, only the platform's ordinary user segment can be considered. This process requires segmenting users, and the specific implementation process is as follows:

[0167] S703: The server stratifies each candidate object in the second candidate object set and uses the candidate object with the target level as the target object.

[0168] The target level can be one or more levels specified based on the expected campaign performance. In this embodiment, to avoid negative impacts on the platform's overall metrics, the selected target level is: the ordinary user level.

[0169] The offline number package generation process in related technologies often relies on task-related prior data, only considering whether the target users of the generated number packages match the content to be delivered. For example, it might recall users interested in certain tags based on the content's related tags to create number packages. However, this method doesn't consider the negative impact of the content on the platform's overall metrics. Typically, the content to be delivered is platform content that needs support; forcibly exposing this type of content to the platform's core users can easily negatively affect their consumption experience and perception, leading to the churn of core users. Furthermore, since the number packages are generated before delivery, it's unknown whether the users in the number packages will actually access the platform. To achieve the target reach, offline number packages usually need to include many times more users than the target number for online delivery.

[0170] To address the aforementioned issues, this application proposes a method of segmenting users based on their behavioral information within a preset historical time period.

[0171] Optionally, the process of stratifying the candidate objects in the second candidate object set can be further divided into the following sub-steps:

[0172] S7031: The server obtains the behavior information of each candidate object in the second candidate object set within a preset historical time period;

[0173] S7032: The server stratifies each candidate object based on its behavioral information.

[0174] The behavioral information of each candidate object within a preset historical time period includes at least the following:

[0175] Number of visits within a preset historical period; total content browsing time within a preset historical period; click-through rate within a preset historical period.

[0176] The number of visits can refer to the number of days a user visits the content creation platform. For example, if a user visits the platform every day, the number of visits increases by 1; or if a user clicks on a page every day, the number of visits increases by 1. The total content browsing time can also be called dwell time or consumption time. In this application, platform users can be segmented based on three metrics: number of days visited, dwell time, and click-through rate.

[0177] Specifically, taking the past seven days from the current time as a preset historical period as an example, for instance, if the current time is 12:00 on November 8, 2021, then the past seven days refer to November 1, 2021 to November 7, 2021.

[0178] For example, daily statistics are compiled using users' spending records from the past seven days on a content creation platform, aggregating each user's spending time, number of days accessed, and click-through rate. The final user segmentation results are roughly as shown in the table below:

[0179] Table 1 User Tiering Standards

[0180]

[0181]

[0182] As shown in Table 1, this application categorizes content creation platform users into 14 levels based on their access days, dwell time, and click-through rate. Users who frequently visit the platform daily but consume very little time per article are classified as web crawlers, corresponding to Level 0. These users are mostly web crawlers, and providing them with supportive content exposure is not beneficial. Levels 1, 2, 4, and 8 are considered cold start users; recalling a large number of these users for phone number packages may prevent the achievement of target user retention. Level 13 users are considered deep users, exhibiting the longest access days and dwell time. Exposing content to Level 13 users can significantly negatively impact overall efficiency metrics. Levels 3, 5, 6, 7, 9, 10, 11, and 12 are considered regular users, used to generate offline phone number packages for each piece of content.

[0183] It should be noted that the click-through rate (CTR) in Table 1 is expressed as "0.XXX" for simplicity, and the actual values ​​may differ. For example, if there were clicks in the past 1-2 days and the total viewing time per week was greater than 20 minutes, the CTR value would be relatively high. Conversely, if there were clicks in the past 5-7 days and the total viewing time per week was 0-5 minutes, the CTR value would be relatively low.

[0184] Based on the embodiments listed above, users can be divided into 14 levels, from Level 0 to Level 13. Taking the target level as the ordinary user level as an example, users in Levels 3, 5, 6, 7, 9, 10, 11, and 12 can be considered as target users.

[0185] Specifically, in the tiered user screening stage, the preliminary pre-offline number packets obtained by the user filtering module can be processed based on the results of user segmentation. Further user filtering will be implemented to retain only users in the general user tier, avoiding contact with the top 5% of deep platform users and users who have just started using the platform.

[0186] In the above implementation, to ensure that the content to be delivered does not harm the overall metrics of the content creation platform (such as user dwell time and completion rate), this strategy adopts a sub-strategy of segmenting users on the content creation platform. Users are divided into 14 tiers based on metrics such as the number of days users visit the platform, dwell time, and click-through rate. During the real-time delivery phase, only ordinary users of the platform are considered, guiding the content to reach only the general consumer users of the content creation platform. This avoids harming the consumption experience of the top 5% of deep users on the content creation platform. At the same time, the user segmentation sub-strategy also ensures that when generating offline number packages, it can effectively avoid including cold-start users who rarely visit the content creation platform, thus ensuring the volume requirements of the delivery business.

[0187] Based on the above implementation method, it can be effectively ensured that the delivered text and image tasks do not affect the platform's newly launched users or active users. This operation ensures the delivery volume target is met and prevents the delivery behavior from having a significant impact on the relevant performance indicators of the content creation platform as a whole.

[0188] It should be noted that content creation platforms prioritize click-through rates and user dwell time in their text and image scenarios, while ad placement focuses more on maintaining a consistent user volume. Therefore, this application primarily uses click-through rate, user dwell time, and number of days visited as metrics for user segmentation in the text and image ad placement scenario of content creation platforms. Furthermore, the user segmentation strategy in this application embodiment can also employ other different metrics, such as conversion rate to followers and like rate, etc., which are not specifically limited here.

[0189] In addition, it is understood that in the specific implementation of this application, user information is involved, including the user's historical behavior, behavior information within a preset time period, and other related data. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data shall comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0190] The following mainly uses the delivery of text and image content as an example. Please refer to [link / reference]. Figure 8 The diagram shown is an implementation flowchart of a text and image delivery task in an embodiment of this application. It specifically includes the following parts: obtaining the content to be delivered, recalling target users, filtering users based on the historical consumption sequence of the task, screening users at different levels through user segmentation, generating number packages, and online delivery.

[0191] Specifically, the process first involves acquiring image and text delivery tasks, and then conducting target user recall for each task to obtain the first candidate object set corresponding to each task. Furthermore, based on the historical consumption sequence corresponding to the acquired task... Perform user filtering to obtain the second candidate object set. Furthermore, by segmenting users and filtering them at different levels, a number package for the target users corresponding to each text and image delivery task can be generated; finally, online delivery can be carried out based on the generated offline number packages.

[0192] The following is combined Figure 8 The flowchart shown provides a detailed explanation of the process for generating number packets for image and text content 1, image and text content 2, ..., image and text content k.

[0193] First, for each piece of text / image content, target user recall is performed using the methods listed above to obtain the first candidate set for each content. Assuming the target audience for these k pieces of text / image content is the same, 100,000, the obtained first candidate set should contain 400,000 users. Then, user filtering is performed on each first candidate set to prevent secondary exposure, resulting in the second candidate set for each piece of text / image content. Assuming that the second candidate set for text / image content 1 still has 300,000 users remaining, text / image content 2 has 350,000 remaining, ..., and text / image content k has 250,000 remaining; next, the users in each second candidate set can be segmented according to the segmentation methods listed in Table 1 above. Suppose that the users in the second candidate set corresponding to image and text content 1 can be divided into two levels: cold start users and ordinary users; the users in the second candidate set corresponding to image and text content 2 can be divided into three levels: crawler users, cold start users, and ordinary users; ..., the users in the second candidate set corresponding to image and text content k can be divided into two levels: deep users and ordinary users. The shaded area represents the ordinary user level. Figure 9 As shown. Furthermore, after filtering users by level, target users can be obtained, and offline number packages can be formed from the target user IDs. For example, number package 1, number package 2, ..., number package k are generated, each containing 200,000 target user IDs.

[0194] In this embodiment, considering that users in the final generated offline number package may not necessarily access the platform, and that the corresponding delivery task will not be exposed to users if they do not access the platform, in order to ensure exposure, if the number of remaining users obtained after the hierarchical user screening stage for a certain content to be delivered is less than twice the target delivery volume, it is necessary to return to the user recall stage to recall more candidate users for generating the delivery user number package.

[0195] One specific implementation is as follows: if the ratio of the number of candidate objects with the target level to the target delivery volume corresponding to a content to be delivered is less than the target value, then a new first candidate object set associated with the target representation features of a content to be delivered is obtained again, and candidate objects with the target level are determined again from the new first candidate object set as target objects, until the ratio of the total number of determined target objects to the target delivery volume is not less than the target value.

[0196] The target value is less than the ratio of the number of candidate objects in the first candidate object set to the target delivery amount.

[0197] Taking the conditions listed above as an example, the target value is 2, and the ratio of the number of candidate objects in the first candidate object set to the target delivery volume is 4. After user filtering and hierarchical user screening, if the number of target hierarchical users in the second candidate object set is less than twice the target delivery volume, then... Figure 8 As shown, return to the target user recall step, obtain a new first candidate object set, and then further perform user filtering and hierarchical user screening steps until the final number of all target users is not less than twice the target delivery volume.

[0198] For example, for content x to be delivered k The corresponding target delivery volume z k When the number is 100,000, the number of users in the first candidate object set is N = 400,000.

[0199] like Figure 10 The diagram illustrates a number package construction process in an embodiment of this application. Assume that in the first round, after target user recall, 400,000 candidate users are obtained. After user filtering, 200,000 remain. After hierarchical user screening, 100,000 target-level candidate users are obtained, i.e., target users. It is determined that 10 / 10 = 1, which is less than the target value of 2. Therefore, a new round of target user recall is needed, again obtaining 400,000 candidate users. After user filtering, 300,000 remain. After hierarchical user screening, 50,000 target users are obtained. It is determined that (10+5) / 10 = 1.5, which is less than the target value of 2. Therefore, a new round of target user recall is needed again. Assume another 50,000 target users are obtained this time. After three rounds, a total of 200,000 target users are obtained. An offline number package can then be generated based on these 200,000 target users.

[0200] It should be noted that, Figure 10 The target user recall scenarios listed are merely illustrative. When conducting the second and third rounds of target user recall, it is not necessary to recall 400,000 candidate users. Adjustments can be made based on the remaining number of target users required. For example, after acquiring 50,000 target users in the first round, there are still 150,000 short of the final 200,000 (denoted as value a), which is 1.5 times the target deployment of 10. In this case, 300,000 can be recalled (denoted as value b), ensuring that the ratio of value b to value a is not less than the target value. No specific limitation is made here.

[0201] In this embodiment of the application, the offline stage is completed for each piece of content x to be delivered. * Once the number package is generated, content can be delivered online. One optional implementation is to carry out the content delivery process in step S34 based on the following process:

[0202] First, the server determines the sequence of content to be delivered for each target object based on the multiple target objects corresponding to each content to be delivered. Then, the server delivers content to the target objects according to the delivery weight of each content to be delivered in the sequence of content to be delivered. The delivery weight represents the degree of relevance between the corresponding content to be delivered and the target object.

[0203] That is, in the online phase, by inverting the offline number packets, a unique identifier is obtained for each user. i The list of content to be delivered is shown below:

[0204]

[0205] Where x1, x2, ..., x q The target audience for these content items all includes user u. i Therefore, the user u obtained through inverted index processing i The list of content to be delivered contains the aforementioned q pieces of content. Furthermore, this list of content to be delivered is in sequence form, that is, the sequence of content to be delivered described above, from x1 to x... q These content items to be deployed and the current user u i The correlation decreases sequentially, meaning x1 represents the correlation with the current user u. i The content to be delivered with the highest relevance. This includes the content to be delivered and the current user's... i The relevance, i.e., the target representation features of the content to be delivered and the user u i The object represents the degree of correlation between features, that is, the similarity between the target vector and the user vector.

[0206] In this embodiment of the application, when user u i When accessing content creation platforms, prioritize displaying tasks that rank higher in the list of content to be published. This can effectively ensure the effectiveness of the content to be published.

[0207] In summary, this application proposes a novel offline number package generation strategy. This strategy recalls relevant target users by combining the content dimension and user behavior dimension of the joint delivery task. Furthermore, it proposes a user segmentation sub-strategy to ensure that the generated offline number packages achieve a balance between the target delivery volume, delivery effectiveness, and impact on overall metrics.

[0208] Based on the same inventive concept, embodiments of this application also provide a content delivery device. For example... Figure 8 As shown, it is a structural schematic diagram of a content delivery device 1100 according to an embodiment of this application, which may include:

[0209] The content acquisition unit 1101 is used to acquire a set of content to be delivered, which includes at least one piece of content to be delivered.

[0210] The first representation unit 1102 is used to obtain the corresponding content representation features based on the content information of each content to be delivered; and to obtain the object representation features of at least one seed object associated with each content to be delivered; the seed object is: a historical object whose sorting result is in the first order range, determined based on the historical behavior of the object in the initial delivery process of the corresponding content to be delivered.

[0211] The second representation unit 1103 is used to determine the target representation features corresponding to each content to be delivered based on the content representation features corresponding to each content to be delivered and at least one object representation feature.

[0212] The content delivery unit 1104 is used to obtain multiple target objects corresponding to each content to be delivered based on each target representation feature, and to deliver each content to be delivered to its corresponding target object.

[0213] Optionally, content delivery unit 1104 is specifically used for:

[0214] For each piece of content to be delivered, perform the following operations:

[0215] Obtain a first set of candidate objects associated with the target representation features of a content to be delivered, and a set of historical objects associated with the content to be delivered. The set of historical objects includes at least one historical object that has viewed a content to be delivered. The object representation features of the candidate objects in the first set are ranked in the second order range with the degree of association between the object representation features and the target representation features.

[0216] Based on the historical object set, the first candidate object set is filtered to obtain the second candidate object set;

[0217] The candidate objects in the second candidate object set are divided into layers, and the candidate objects with the target layer are taken as the target objects.

[0218] Optionally, content delivery unit 1104 is specifically used for:

[0219] Compare the set of historical objects with the set of first candidate objects;

[0220] Based on the comparison results, the candidate objects located in the historical object set are deleted from the first candidate object set to obtain the second candidate object set.

[0221] Optional, content delivery unit 1104:

[0222] Obtain the behavior information of each candidate object in the second candidate object set within a preset historical time period;

[0223] Based on the behavioral information of the candidate objects, each candidate object is stratified.

[0224] Optionally, the behavioral information of each candidate object within a preset historical time period shall include at least:

[0225] Number of visits within a preset historical time period;

[0226] Total browsing time within the preset historical time period;

[0227] Click-through rate within a preset historical time period.

[0228] Optionally, the content delivery unit 1104 is also used for:

[0229] If the ratio of the number of candidate objects with target level to the target delivery volume corresponding to a content to be delivered is less than the target value, then a new first candidate object set associated with the target representation features of a content to be delivered is obtained again, and candidate objects with target level are determined again from the new first candidate object set as target objects, until the ratio of the total number of determined target objects to the target delivery volume is not less than the target value.

[0230] The target value is less than the ratio of the number of candidate objects in the first candidate object set to the target delivery amount.

[0231] Optionally, the second representation unit 1103 is specifically used for:

[0232] Based on at least one object representation feature corresponding to each content to be delivered, determine the behavioral representation feature corresponding to each content to be delivered.

[0233] Based on the content representation features and behavioral representation features of each content to be delivered, the target representation features corresponding to each content to be delivered are determined.

[0234] Optionally, the second representation unit 1103 is specifically used for:

[0235] Linear interpolation is performed on the content representation features and behavior representation features corresponding to each content to be delivered to obtain the target representation features corresponding to each content to be delivered.

[0236] Optionally, content delivery unit 1104 is specifically used for:

[0237] Based on the multiple target objects corresponding to each content to be delivered, determine the sequence of content to be delivered for each target object;

[0238] Based on the respective placement weights of each content in the content sequence to be placed, content is placed on the target audience. The placement weights represent the degree of relevance between the corresponding content and the target audience.

[0239] In this embodiment, when determining the target audience for the content to be delivered, both the content dimension and the behavior dimension are combined. Specifically, from the content dimension, the content representation features of each content to be delivered are obtained. From the behavior dimension, based on the relevant information of each content to be delivered during the initial delivery process, the object representation features of at least one associated seed object are determined. Then, the content dimension and the behavior dimension are combined, and the content representation features and object representation features are combined to determine the target representation features corresponding to each content to be delivered. Finally, the target representation features are used to select matching target objects as delivery targets. This approach, by jointly recalling target objects from both the content dimension (prior knowledge) and the behavior dimension (posterior knowledge) for subsequent content delivery, can effectively improve the accuracy of target object selection and increase content delivery efficiency.

[0240] For ease of description, the above sections are divided into modules (or units) according to their functions and described separately. Of course, in implementing this application, the functions of each module (or unit) can be implemented in one or more software or hardware components.

[0241] Having introduced the content delivery method and apparatus according to exemplary embodiments of this application, we will now introduce another exemplary embodiment of the content delivery apparatus according to this application.

[0242] Those skilled in the art will understand that various aspects of this application can be implemented as a system, method, or program product. Therefore, various aspects of this application can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or a combination of hardware and software implementations, collectively referred to herein as a "circuit," "module," or "system."

[0243] Based on the same inventive concept as the above-described method embodiments, this application also provides an electronic device. In one embodiment, the electronic device may be a server, such as... Figure 2 The server 220 is shown. In this embodiment, the structure of the electronic device can be as follows: Figure 12 As shown, it includes a memory 1201, a communication module 1203, and one or more processors 1202.

[0244] The memory 1201 is used to store computer programs executed by the processor 1202. The memory 1201 may mainly include a program storage area and a data storage area. The program storage area may store the operating system and programs required to run instant messaging functions, etc.; the data storage area may store various instant messaging information and operation instruction sets, etc.

[0245] Memory 1201 may be volatile memory, such as random-access memory (RAM); memory 1201 may also be non-volatile memory, such as read-only memory, flash memory, hard disk drive (HDD), or solid-state drive (SSD); or memory 1201 may be any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto. Memory 1201 may be a combination of the above-described memories.

[0246] The processor 1202 may include one or more central processing units (CPUs) or digital processing units, etc. The processor 1202 is used to implement the above-described content delivery method when calling a computer program stored in the memory 1201.

[0247] The communication module 1203 is used to communicate with terminal devices and other servers.

[0248] This application embodiment does not limit the specific connection medium between the memory 1201, communication module 1203, and processor 1202. This application embodiment... Figure 12 The memory 1201 and the processor 1202 are connected via a bus 1204, and the bus 1204 is in Figure 12 The diagram uses thick lines to describe the connections between other components; these are for illustrative purposes only and should not be considered limiting. The 1204 bus can be divided into address bus, data bus, control bus, etc. For ease of description, Figure 12 It is described using only a thick line, but does not indicate that there is only one bus or one type of bus.

[0249] The memory 1201 stores a computer storage medium, which stores computer-executable instructions for implementing the content delivery method of this application embodiment. The processor 1202 is used to execute the above-described content delivery method, such as... Figure 3 As shown.

[0250] In another embodiment, the electronic device can also be other electronic devices, such as... Figure 2 The terminal device 210 is shown. In this embodiment, the electronic device can be structured as follows: Figure 13 As shown, it includes components such as: communication component 1310, memory 1320, display unit 1330, camera 1340, sensor 1350, audio circuit 1360, Bluetooth module 1370, processor 1380, etc.

[0251] The communication component 1310 is used to communicate with the server. In some embodiments, it may include a Circuit-Wireless Fidelity (WiFi) module, which is a short-range wireless transmission technology. Electronic devices can use the WiFi module to help users send and receive information.

[0252] The memory 1320 can be used to store software programs and data. The processor 1380 executes various functions of the terminal device 210 and performs data processing by running the software programs or data stored in the memory 1320. The memory 1320 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. The memory 1320 stores an operating system that enables the terminal device 210 to run. In this application, the memory 1320 may store the operating system and various applications, and may also store code that executes the content delivery method of the embodiments of this application.

[0253] The display unit 1330 can also be used to display information input by the user or information provided to the user, as well as various menus of the terminal device 210, in a graphical user interface (GUI). Specifically, the display unit 1330 may include a display screen 1332 disposed on the front of the terminal device 210. The display screen 1332 may be configured as a liquid crystal display, a light-emitting diode, or the like. The display unit 1330 can be used to display client-related operation interfaces, content to be displayed, etc., as described in this embodiment.

[0254] The display unit 1330 can also be used to receive input digital or character information and generate signal inputs related to user settings and function control of the terminal device 210. Specifically, the display unit 1330 may include a touch screen 1331 disposed on the front of the terminal device 210, which can collect touch operations of the user on or near it, such as clicking a button, dragging a scroll box, etc.

[0255] The touchscreen 1331 can be placed over the display screen 1332, or the touchscreen 1331 and the display screen 1332 can be integrated to realize the input and output functions of the terminal device 210. After integration, it can be referred to as a touch display screen. In this application, the display unit 1330 can display the application program and the corresponding operation steps.

[0256] Camera 1340 can be used to capture still images, which users can then post comments on via the application. There can be one or multiple cameras 1340. An object is projected onto a photosensitive element through a lens, generating an optical image. This photosensitive element can be a charge-coupled device (CCD) or a complementary metal-oxide-semiconductor (CMOS) phototransistor. The photosensitive element converts the light signal into an electrical signal, which is then transmitted to the processor 1380 for conversion into a digital image signal.

[0257] The terminal device may also include at least one sensor 1350, such as an accelerometer 1351, a proximity sensor 1352, a fingerprint sensor 1353, and a temperature sensor 1354. The terminal device may also be equipped with other sensors such as a gyroscope, barometer, hygrometer, thermometer, infrared sensor, light sensor, and motion sensor.

[0258] Audio circuitry 1360, speaker 1361, and microphone 1362 provide an audio interface between the user and terminal device 210. Audio circuitry 1360 converts received audio data into electrical signals, which are then transmitted to speaker 1361, where they are converted into sound signals for output. Terminal device 210 may also be equipped with volume buttons for adjusting the volume of the sound signal. Conversely, microphone 1362 converts collected sound signals into electrical signals, which are then received by audio circuitry 1360, converted into audio data, and output to communication component 1310 for transmission to, for example, another terminal device 210, or to memory 1320 for further processing.

[0259] The Bluetooth module 1370 is used to interact with other Bluetooth devices that also have a Bluetooth module via the Bluetooth protocol. For example, a terminal device can establish a Bluetooth connection with a wearable electronic device (such as a smartwatch) that also has a Bluetooth module through the Bluetooth module 1370, thereby exchanging data.

[0260] The processor 1380 is the control center of the terminal device, connecting various parts of the terminal through various interfaces and lines. It executes various functions and processes data by running or executing software programs stored in the memory 1320 and calling data stored in the memory 1320. In some embodiments, the processor 1380 may include one or more processing units; the processor 1380 may also integrate an application processor and a baseband processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the baseband processor mainly handles wireless communication. It is understood that the baseband processor may not be integrated into the processor 1380. In this application, the processor 1380 can run the operating system, applications, user interface display and touch response, and the content delivery method of this embodiment. Furthermore, the processor 1380 is coupled to the display unit 1330.

[0261] In some possible implementations, various aspects of the content delivery method provided in this application can also be implemented as a program product, which includes program code. When the program product is run on an electronic device, the program code is used to cause the electronic device to perform the steps of the content delivery method according to the various exemplary embodiments of this application described above. For example, the electronic device can perform actions such as... Figure 3 The steps are shown in the figure.

[0262] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0263] The program product of the embodiments of this application may employ a portable compact disc read-only memory (CD-ROM) and include program code, and may run on a computing device. However, the program product of this application is not limited thereto. In this document, the readable storage medium may be any tangible medium that contains or stores a program that may be used by or in conjunction with a command execution system, apparatus, or device.

[0264] A readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying readable program code. This propagated data signal may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting a program for use by or in conjunction with a command execution system, apparatus, or device.

[0265] The program code contained on the readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0266] Program code for performing the operations of this application can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0267] It should be noted that although several units or sub-units of the device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of this application, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided and embodied by multiple units.

[0268] Furthermore, although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0269] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0270] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, produce a machine for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0271] These computer program commands may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the commands stored in the computer-readable storage medium produce an article of manufacture including command means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0272] These computer program commands can also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing the commands executed on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0273] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0274] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A content delivery method, characterized in that, The method includes: Obtain a set of content to be delivered, wherein the set of content to be delivered includes at least one piece of content to be delivered; Based on the content information of each content to be delivered, obtain the corresponding content representation features; and obtain the object representation features of at least one seed object associated with each content to be delivered; the seed object is a historical object whose sorting result is in the first order range, determined based on the historical behavior of the object in the initial delivery process of the corresponding content to be delivered. Based on the content representation features corresponding to each content to be delivered, and at least one object representation feature, the target representation features corresponding to each content to be delivered are determined respectively; Based on each target representation feature, multiple target objects corresponding to each content to be delivered are obtained, and each content to be delivered is delivered to its corresponding target object. The target object corresponding to each content to be delivered is: a candidate object with a target level in a first candidate object set associated with the target representation feature of the content to be delivered; wherein, the ratio of the total number of candidate objects with a target level to the target delivery amount corresponding to the content to be delivered is not less than a target value, and the target value is less than the ratio of the number of candidate objects in the first candidate object set to the target delivery amount.

2. The method as described in claim 1, characterized in that, The step of obtaining multiple target objects corresponding to each content to be delivered based on each target representation feature includes: For each piece of content to be delivered, perform the following operations: Obtain a first set of candidate objects associated with the target representation features of a content to be delivered, and a set of historical objects associated with the content to be delivered, the set of historical objects including at least one historical object that has viewed the content to be delivered, and the object representation features of the candidate objects in the first set of candidate objects are ranked in the second order range with the correlation between the target representation features and the object representation features. Based on the historical object set, the first candidate object set is filtered to obtain the second candidate object set; The candidate objects in the second candidate object set are divided into layers, and the candidate objects with the target layer are taken as the target objects.

3. The method as described in claim 2, characterized in that, The step of filtering the first candidate object set based on the historical object set to obtain a second candidate object set includes: The set of historical objects is compared with the first set of candidate objects; Based on the comparison results, candidate objects located in the historical object set are deleted from the first candidate object set to obtain the second candidate object set.

4. The method as described in claim 2, characterized in that, The step of stratifying the candidate objects in the second candidate object set includes: Obtain the behavior information of each candidate object in the second candidate object set within a preset historical time period; Based on the behavioral information of the candidate objects, the candidate objects are stratified.

5. The method as described in claim 4, characterized in that, The behavioral information of each candidate within a preset historical time period includes at least: The number of visits within the preset historical time period; Total content browsing time within the preset historical time period; Click-through rate within the preset historical time period.

6. The method as described in claim 2, characterized in that, The method further includes: If the ratio of the number of candidate objects with the target level to the target delivery volume corresponding to the content to be delivered is less than the target value, then a new first candidate object set associated with the target representation features of the content to be delivered is obtained again, and candidate objects with the target level are determined again from the new first candidate object set as target objects, until the ratio of the total number of determined target objects to the target delivery volume is not less than the target value.

7. The method as described in claim 1, characterized in that, The step of determining the target representation features corresponding to each content to be delivered, based on the content representation features corresponding to each content to be delivered and at least one object representation feature, includes: Based on at least one object representation feature corresponding to each of the content to be delivered, determine the behavioral representation feature corresponding to each of the content to be delivered; Based on the content representation features and behavior representation features corresponding to each of the content to be delivered, the target representation features corresponding to each of the content to be delivered are determined.

8. The method as described in claim 7, characterized in that, The step of determining the target representation features corresponding to each of the content to be delivered based on its respective content representation features and behavioral representation features includes: Linear interpolation is performed on the content representation features and behavior representation features corresponding to each content to be delivered to obtain the target representation features corresponding to each content to be delivered.

9. The method according to any one of claims 1 to 8, characterized in that, The step of delivering each piece of content to its corresponding target audience includes: Based on the multiple target objects corresponding to each of the content to be delivered, a sequence of content to be delivered corresponding to each target object is determined. Based on the delivery weight of each content in the sequence of content to be delivered, content is delivered to the target object, whereby the delivery weight represents the degree of correlation between the corresponding content to be delivered and the target object.

10. A content delivery device, characterized in that, include: A content acquisition unit is used to acquire a set of content to be delivered, wherein the set of content to be delivered includes at least one piece of content to be delivered. The first representation unit is used to obtain the corresponding content representation features of each content to be delivered based on the content information of each content to be delivered. In addition, the object representation features of at least one seed object associated with each content to be delivered are obtained respectively; The seed object is a historical object whose sorting result is within the first order range, determined based on the object's historical behavior during the initial delivery process of the corresponding content to be delivered. The second representation unit is used to determine the target representation features corresponding to each content to be delivered based on the content representation features corresponding to each content to be delivered, and at least one object representation feature. The content delivery unit is used to obtain multiple target objects corresponding to each content to be delivered based on each target representation feature, and to deliver each content to be delivered to its corresponding target object. The target object corresponding to each content to be delivered is a candidate object with a target level in a first candidate object set associated with the target representation feature of the content to be delivered. The ratio of the total number of candidate objects with a target level to the target delivery amount corresponding to the content to be delivered is not less than a target value, and the target value is less than the ratio of the number of candidate objects in the first candidate object set to the target delivery amount.

11. The apparatus as claimed in claim 10, characterized in that, The content delivery unit is specifically used for: For each piece of content to be delivered, perform the following operations: Obtain a first set of candidate objects associated with the target representation features of a content to be delivered, and a set of historical objects associated with the content to be delivered, the set of historical objects including at least one historical object that has viewed the content to be delivered, and the object representation features of the candidate objects in the first set of candidate objects are ranked in the second order range with the correlation between the target representation features and the object representation features. Based on the historical object set, the first candidate object set is filtered to obtain the second candidate object set; The candidate objects in the second candidate object set are divided into layers, and the candidate objects with the target layer are taken as the target objects.

12. The apparatus as claimed in claim 11, characterized in that, The content delivery unit is specifically used for: The set of historical objects is compared with the first set of candidate objects; Based on the comparison results, candidate objects located in the historical object set are deleted from the first candidate object set to obtain the second candidate object set.

13. An electronic device, characterized in that, It includes a processor and a memory, wherein the memory stores program code that, when executed by the processor, causes the processor to perform the steps of any of the methods described in claims 1 to 9.

14. A computer-readable storage medium, characterized in that, It includes program code that, when run on an electronic device, causes the electronic device to perform the steps of any of the methods described in claims 1 to 9.

15. A computer program product, characterized in that, The method includes computer instructions stored in a computer-readable storage medium; when a processor of an electronic device reads the computer instructions from the computer-readable storage medium, the processor executes the computer instructions, causing the electronic device to perform the steps of any one of claims 1 to 9.

Images