Content recommendation method, apparatus, device, and medium

Abstract

This application discloses a content recommendation method, apparatus, device, and medium, applicable to the field of computer technology. The method includes: acquiring at least one content publishing account that has an interactive relationship with a target account within a historical time period; acquiring first published content from the published content of each content publishing account as a first candidate content set; acquiring at least one similar publishing account with similar publishing attributes to each content publishing account; acquiring target similar publishing accounts based on the account exposure of the similar publishing accounts; acquiring second published content from the published content of the target similar publishing accounts as a second candidate content set; and recommending the published content in the target content set to the target account when constructing a target content set using the first and second candidate content sets. Using this application embodiment can improve the content recommendation effect.

Description

Technical Field

[0001] This application relates to the field of Internet technology, specifically to the field of data display technology, and in particular to a content recommendation method, apparatus, device, and medium. Background Technology

[0002] Currently, content platforms can recommend content to users. For example, they can recommend published content on the homepage. In this process, content can be selected from the published content library that the user might be interested in, such as predicting the user's level of interest in each piece of content to filter the recommended content. However, this content recommendation method is relatively simplistic. Directly filtering content from the published content library relies solely on the direct association between the user and the published content, failing to consider richer dimensions for content selection. This results in a limited range of recommended content and poor recommendation effectiveness. Summary of the Invention

[0003] This application provides a content recommendation method, apparatus, device, and medium that can enhance the diversity of content recommendation methods, thereby improving the richness of recommended content and the effectiveness of content recommendation.

[0004] On the one hand, embodiments of this application provide a content recommendation method, which includes: Obtain at least one content publishing account that has an interactive relationship with the target account during the historical period, and extract the first published content that meets the quality threshold from the published content of each content publishing account as the first candidate content set; Obtain at least one similar publishing account that has similar publishing attributes to each content publishing account. Based on the account exposure of the similar publishing account, obtain the target similar publishing account from the at least one similar publishing account. Then, obtain the second publishing content with a quality threshold from the publishing content published by the target similar publishing account, as the second candidate content set. When constructing the target content set using the first and second candidate content sets, the published content from the target content set is recommended to the target account.

[0005] On one hand, embodiments of this application provide a content recommendation device, which includes: The content acquisition module is used to acquire at least one content publishing account that has an interactive relationship with the target account during the historical period, and to acquire the first published content that meets the quality threshold from the published content of each content publishing account, as the first candidate content set; The content acquisition module is also used to acquire at least one similar publishing account that has similar publishing attributes to each content publishing account, acquire a target similar publishing account from at least one similar publishing account based on the account exposure of the similar publishing account, and acquire second publishing content with a quality threshold from the publishing content published by the target similar publishing account as a second candidate content set; The content recommendation module is used to recommend published content from the target content set to the target account when the target content set is constructed using the first candidate content set and the second candidate content set.

[0006] On one hand, embodiments of this application provide an electronic device including a processor and a memory, wherein the memory is used to store a computer program, the computer program including program instructions, and the processor is configured to invoke the program instructions to execute some or all of the steps in the above method.

[0007] On one hand, embodiments of this application provide a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, are used to perform some or all of the steps in the above-described method.

[0008] Accordingly, according to one aspect of this application, a computer program product or computer program is provided, which includes computer instructions that, when executed by a processor, can implement some or all of the steps in the above-described method.

[0009] In this embodiment, at least one content publishing account that has an interactive relationship with the target account during a historical period can be obtained. That is, by account, content publishing accounts that are closely related to the target account can be obtained based on the target account's behavior. This indicates that the target account is more likely to be interested in the content published by such accounts. Therefore, the first published content that meets the quality threshold can be obtained from the published content of each content publishing account, i.e., high-quality content published by authors with high affinity to the target account is selected as the first candidate content set. Furthermore, to enrich the source of the selected content, at least one similar publishing account with similar publishing attributes to each content publishing account can also be obtained. The published content published by authors similar to the high-affinity authors may also be published content that users are interested in. Therefore, similar publishing accounts can be used as a basis for further selection. The method involves obtaining target similar publishing accounts from at least one similar publishing account, and then extracting second-tier publishing content that meets a quality threshold from the content published by the target similar publishing accounts. This second-tier content set can be used to filter out content published by similar authors. The target content set can then be constructed using the first and second-tier content sets to achieve content recommendation. Compared to directly determining a user's interest in each published content from the published content library, the proposed method allows for content filtering from richer dimensions, such as filtering content to be recommended based on highly intimate exclusive accounts or similar accounts. This makes it more likely to identify published content that users are interested in, thereby increasing the diversity of content recommendation methods and improving the richness and effectiveness of recommended content. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 A schematic diagram of a network architecture provided for an embodiment of this application; Figure 2 A schematic diagram illustrating a content recommendation scenario provided in an embodiment of this application; Figure 3 A flowchart illustrating a content recommendation method provided in this application embodiment. Figure 1 ; Figures 4-5 A schematic diagram illustrating a content recommendation scenario provided in an embodiment of this application; Figure 6A flowchart illustrating a content recommendation method provided in this application embodiment. Figure 2 ; Figure 7 A schematic diagram illustrating a recommended scenario for published content, provided as an embodiment of this application; Figure 8 This is a schematic diagram of the structure of a content recommendation device provided in an embodiment of this application; Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0012] The content recommendation method proposed in this application is implemented in an electronic device, which can be a server or a terminal. The server can be a standalone 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 computing, cloud functions, cloud storage, network services, cloud communication, middleware services, and big data and artificial intelligence platforms. The terminal can be a smartphone, tablet computer, laptop computer, desktop computer, etc., but is not limited to these.

[0013] In the description of the embodiments of this application, the published content may refer to notes (including text and images), short videos, medium-length videos, etc., posted by users on content platforms. Alternatively, it may also be live streams, instantaneous content (a short content that is shared and exists in real time), points of interest (such as group chats, live streams, etc.), routes (such as cycling routes, travel guides, etc.). The specific content information included in the published content applicable to different scenarios can be adjusted accordingly, and the content information in the published content can be determined by the publisher (such as the target audience). The specific type of published content is not limited here.

[0014] One of the network architecture diagrams proposed in the content recommendation method is as follows: Figure 1 As shown, the network architecture may include server 100 (the number of servers is not limited) and a cluster of terminal devices (the number of terminal devices is not limited, such as terminal device 200a, terminal device 200b, ..., terminal device 200n), wherein communication connections may exist between servers. Simultaneously, a server may have a communication connection with any terminal device, so that the server can interact with the terminal device through this communication connection. The aforementioned communication connection is not limited in method; it can be a direct or indirect connection via wired communication, a direct or indirect connection via wireless communication, or other methods, which are not limited herein. Furthermore, it is understood that the electronic devices involved in the embodiments of this application may be... Figure 1 The terminal device shown can also be Figure 1The server shown.

[0015] It should be understood that, such as Figure 1 Each terminal device in the shown cluster can have an application client installed for content recommendation. This application client can be of any type, such as a social networking client, instant messaging client (e.g., a conferencing client), entertainment client (e.g., a live streaming client), multimedia client (e.g., a video client), news client (e.g., a news and information client), shopping client, or any other client capable of displaying text, images, audio, and video data. No specific type of application client is limited here.

[0016] For example, an application client refers to a client that can send and receive internet messages instantly and has information functions. Specifically, the target account's terminal device (such as...) Figure 1 When the terminal device 200a shown starts the application client and enters the application homepage, it can request the published content to be displayed on the application homepage from the server. At this time, the server can build a target content set for the target account and recommend the published content in the target content set to the target account.

[0017] Optionally, the aforementioned terminal devices and servers can be logically separated. Therefore, when referring to terminal devices and servers below, they may be physically the same device or different devices.

[0018] For further information, please refer to [link / reference]. Figure 2 , Figure 2 This is a schematic diagram illustrating a content recommendation scenario provided in an embodiment of this application. When it is necessary to recommend published content to a target account, at least one content publishing account (such as account a21, account a22, and account a23) with interactive relationships with the target account within a historical time period can be obtained. This content publishing account refers to an account that has published content, and the interactive relationships are used to indicate that the target account may be interested in the published content by that content publishing account.

[0019] Therefore, the first published content that meets the quality threshold can be selected from the published content of each content publishing account and used as the first candidate content set for recommendation.

[0020] Specifically, at least one similar publishing account with similar publishing attributes to each content publishing account can be obtained. In other words, accounts similar to the content publishing account are potential accounts that can interact and be associated with the target account. Therefore, the target account may be interested in the content published by similar publishing accounts.

[0021] Therefore, a target similar posting account (such as account b21, account b22, account b23) can be obtained from at least one similar posting account based on the account exposure of similar posting accounts. That is, the target account is more likely to be interested in similar posting accounts with higher attention, and the content posted by the target similar posting account is also more valuable.

[0022] Among them, second-generation content that meets the quality threshold is obtained from the content published by accounts with similar targets, and this second set of recommended candidate content is selected.

[0023] At this point, a target content set can be constructed using the first and second candidate content sets, and content recommendations can be made for the target account using the target content set.

[0024] Therefore, multiple methods can be used to filter recommended candidate content, making the sources of recommended content more diverse and more likely to recommend content that is of interest to target accounts. This can improve the effectiveness of content recommendation, increase user interaction and retention on the content platform, and also improve the content publishing effect of accounts with similar targets. In other words, it can submit higher-quality content to more interested accounts, thereby helping high-quality accounts gain exposure on the content platform and enabling more accounts to interact with them (such as following high-quality accounts and browsing their published content).

[0025] In specific embodiments of this application, scenarios involving the acquisition of user information and related data, such as the acquisition of user information (e.g., published content), require user permission or consent. That is, when these embodiments are applied to specific products or technologies, the collection, use, and processing of relevant user data comply with the relevant laws, regulations, and standards of the relevant regions. For example, interactive pages can be used to provide prompts indicating which data will be collected or acquired. Specifically, lists or other methods can be used to present the types and content of this data to the user. Data collection and processing will only proceed after a confirmation or instruction to allow data collection is received on the interactive page.

[0026] The scenarios described above are merely examples and do not constitute a limitation on the application scenarios of the technical solutions provided in this application. The technical solutions of this application can also be applied to other scenarios. For example, as those skilled in the art will know, with the evolution of system architecture and the emergence of new business scenarios, the technical solutions provided in this application are also applicable to similar technical problems.

[0027] Based on the above description, this application proposes a content recommendation method, which can be executed by the aforementioned electronic device, specifically... Figure 1 The server shown. Please refer to [link / reference]. Figure 3 , Figure 3 A flowchart illustrating a content recommendation method provided in this application embodiment. Figure 1 .like Figure 3 As shown, the content recommendation method of this application embodiment may include the following flow: S101. Obtain at least one content publishing account that has an interactive relationship with the target account during the historical period, and obtain the first published content whose content quality reaches the quality threshold from the published content published by each content publishing account, as the first candidate content set.

[0028] The target account can be any account on the content platform or a specific type of account. For example, a target account refers to at least one of the following: an account that registers for the target application for the first time within a preset time period; an account that downloads and logs into the target application (content platform) on a terminal device, then uninstalls the target application and re-downloads and logs into the target application within a preset time period.

[0029] The target account can refer to a new user (such as a user who registered within 30 days and has a small amount of interaction on the content platform) or a user who has uninstalled and returned (such as an account that uninstalled the content platform within a period of time and then re-downloaded and logged in). Both new users and returning users refer to those who have a small amount of interaction on the content platform (such as following other accounts or viewing content posted by other accounts). Based on this small amount of interaction, we can identify high-intimacy accounts of the target account and recommend their content. This can not only attract the target account to interact more on the content platform, thereby improving the target account retention rate, but also help increase the exposure of the recommended accounts (that is, recommend the content posted by these accounts to accounts that are more interested in them).

[0030] For example, when it is necessary to recommend content to a target account, the server can receive a content recommendation request sent by the content platform on the terminal device. Based on the content recommendation request, the server executes the technical solution of this application to obtain the published content recommended to the target account.

[0031] Interactive association refers to accounts that have a high degree of intimacy with the target account, i.e., high intimacy accounts. This means that the target account is more likely to be interested in the content posted by such accounts.

[0032] Among them, the at least one content publishing account refers to an account with a high degree of intimacy with the target account within a certain historical period, such as a content publishing account with interactive relationships with the target account within 30 days. In other words, content recommendations can be achieved through accounts that the target account has recently been interested in.

[0033] The method for acquiring content publishing accounts can be determined statistically (i.e., a white-box screening strategy). For example, a pool of candidate publishing accounts can be selected first, then the interaction data of the target account with respect to these candidate accounts can be statistically analyzed (e.g., the number of posts liked, the duration of likes, etc.). The account intimacy between the target account and the candidate accounts can then be calculated statistically. For instance, a calculation formula can be configured to calculate account intimacy. For example, the like time period can be determined, with different like time periods having different weight parameters (e.g., the longer the like time period, the lower the weight parameter). This allows for a weighted sum of the number of posts liked in different like time periods, and the weighted result can be used as the account intimacy. The specific type and scenario of the interaction data are not limited here; it can be interaction data from various social scenarios (e.g., interaction data related to published content, interaction data related to live streams, interaction data related to products, etc.). Furthermore, since different social scenarios and interaction data exist, appropriate calculation formulas can be configured to calculate the account intimacy between the target account and the candidate publishing accounts.

[0034] Alternatively, content publishing accounts can be obtained through model prediction (which can be understood as a black-box screening strategy). For example, the account feature data of the target account and the account feature data of the candidate publishing accounts can be obtained, and the model can predict and output the account affinity based on the account feature data of the target account and the account feature data of the candidate publishing accounts.

[0035] There are no restrictions on the methods for obtaining content publishing accounts; any combination of one or more of the methods mentioned above is acceptable.

[0036] Taking a white-box screening strategy as an example, the method for obtaining content publishing accounts can be as follows: First-line candidate publishing accounts that have interacted with the target account; account interaction includes at least one of the following: following behavior, chat behavior. Based on the strength of the account interaction (i.e., account intimacy) between the target account and the first-line candidate publishing accounts, at least one content publishing account is determined from the first-line candidate publishing accounts. For example, accounts followed by the target account or accounts that have private chat messages with the target account can be obtained.

[0037] Account interactions can also include multiple likes (e.g., liking content posted by a certain account multiple times; the account that has liked the content more than 5 times can be considered the first candidate for posting content), multiple favorites (e.g., favorites posted by a certain account multiple times; the account that has favorited the content more than 5 times can be considered the first candidate for posting content), and joining groups (e.g., joining a group chat created by a certain account; the group owner of the group chat that the target account has joined can be considered the first candidate for posting content).

[0038] The specific behavior of account interaction is not limited here. This account interaction is used to indicate that the target account may be interested in the content published by the first candidate publishing account (rather than being interested in only a single post).

[0039] Among them, the account interaction strength is the interaction strength between the target account and the first candidate posting account. The higher the account interaction strength, the more the target account interacts with the first candidate posting account, that is, the higher the account intimacy between the target account and the first candidate posting account.

[0040] Determining the interaction strength between the target account and the first candidate posting account can be achieved by obtaining the interaction data (such as interaction data within 30 days) that the target account has performed against the first candidate posting account in N (N is a positive integer) social scenarios within a historical period, and determining the interaction strength based on the interaction data in the N social scenarios.

[0041] Here, N social scenarios can be different types of scenarios on a content platform. For example, interaction data in a recommendation scenario refers to data obtained from the interactive behaviors performed by the target account on the content recommendation page (such as the application homepage) towards the first candidate publishing account. For example, the time when the target account clicks on the content published by the first candidate publishing account on the content recommendation page, or the time when the target account likes the content published by the first candidate publishing account on the content recommendation page.

[0042] For example, social scenarios can also include live streaming scenarios, which refer to social pages related to live streaming (such as a live streaming page that is currently live, a live streaming list page that displays live streaming cards, etc.). Interaction data in live streaming scenarios can refer to the target account's viewing time on the live streaming page of the first candidate posting account, the number of items purchased on the live streaming page of the posted live stream, the number of live streams viewed by the first candidate posting account, etc.

[0043] For example, social scenarios can also include chat scenarios, which refer to the chat interaction between the target account and the first candidate posting account on the private message page. Interaction data in chat scenarios can refer to the frequency, number of private messages sent by the target account to the first candidate posting account, or the time of the most recent private message sent, etc.

[0044] For example, social scenarios can also include content interaction scenarios, which refer to the scenario where the target account is viewing the content published by the first candidate publishing account. In the content interaction scenario, the interaction data can refer to the number of posts liked or the time of liking (e.g., how many posts liked and the time of each like), the number of posts saved or the time of saving (e.g., how many posts saved and the time of each saving), the number of posts commented on or the time of commenting (e.g., how many posts commented and the time of each comment), and so on.

[0045] For example, social scenarios can also include account interaction scenarios, which refer to scenarios where users view and interact with a target account. Interaction data in account interaction scenarios can refer to the number of times the user visits the account page of the first candidate posting account and the duration of each visit, or the duration of following the target account, the number of times the user shares the first candidate posting account through its account page, and the duration of such sharing, etc.

[0046] This document does not specify any limitations on social scenarios or specific interaction data.

[0047] Therefore, we can statistically analyze the effective exposure, effective reach, and interaction data of target accounts towards the first candidate posting account in various social scenarios (such as recommendation scenarios, live streaming scenarios, chat scenarios, content interaction scenarios, etc.) within historical periods (such as the last 7 days or 30 days). Then, we can calculate the scene intimacy (i.e. scene interaction intensity, which represents the intimacy between the target account and the first candidate posting account in a social scenario) in different social scenarios. By weighting, we can synergistically integrate different social scenarios to obtain the specific scene intimacy reflecting a certain period of time and the global intimacy (i.e., account interaction intensity) obtained from the scene intimacy.

[0048] Therefore, the interaction strength between the target account and the first candidate posting account in each of the N social scenarios can be determined based on the interaction data in each scenario. The account interaction strength between the target account and the first candidate posting account can then be determined based on the interaction strength across the N social scenarios. For example, the weighted sum of the interaction strengths across the N social scenarios can be used to obtain the account interaction strength. The weighting coefficients for different social scenarios can be set by the relevant business personnel.

[0049] Taking a social scenario as an example, determining the intensity of interaction between the target account and the first candidate posting account in a social context can be achieved by weighted summing of the number of interactions and the duration of interactions. For instance, in a recommendation scenario, the interaction data includes the number of times and the duration during which the target account clicks on content posted by the first candidate posting account on the content recommendation page.

[0050] Specifically, the number of times the target account clicks on the content published by the first candidate publishing account within different time periods can be determined (e.g., every 3 days is considered a time period). Different time periods have different weighting coefficients. Based on the weighting coefficients, the corresponding quantities for different time periods are weighted and summed to obtain the scene interaction intensity in the recommendation scenario.

[0051] For example, in the context of social interaction, the interaction data includes the number of posts liked or the time of liking, and the number of posts saved or the time of saving.

[0052] Specifically, the number of posts published by the first candidate posting account that the target account liked within different time periods (e.g., 3 days per time period) can be determined. The number of likes corresponding to different time periods is weighted and summed based on a weighting coefficient to obtain a first weighted value. The number of posts published by the first candidate posting account that the target account collected within different time periods (e.g., 3 days per time period) can be determined. The number of collections corresponding to different time periods is weighted and summed based on a weighting coefficient to obtain a second weighted value. The sum of the first weighted value and the second weighted value can be used as the scene interaction intensity in the content interaction scenario.

[0053] The specific method for obtaining the corresponding scene interaction intensity based on interaction data statistics for different social scenarios is not limited and can be configured by relevant business personnel.

[0054] Optionally, determining the intensity of account interaction based on interaction data in different social scenarios can also be achieved through prediction using network models. For example, an interaction intensity processing model (such as a transform network model or a large language model) can be built to predict the intensity of account interaction. Interaction data from social scenarios can be input into the interaction intensity processing model to predict the corresponding account interaction intensity.

[0055] Specifically, determining at least one content publishing account from the first candidate publishing accounts can be achieved by selecting first candidate publishing accounts with an interaction intensity greater than an interaction intensity threshold as at least one content publishing account that has an interactive relationship with the target account. Alternatively, it can be achieved by selecting a specified number of first candidate publishing accounts with the highest interaction intensity as at least one content publishing account.

[0056] Taking the black-box screening strategy as an example, the way to obtain content publishing accounts can be: based on the account selection strategy, select a second candidate publishing account that has a potential relationship with the target account from the account set. That is, the second candidate publishing account is an account that may have an interactive relationship. At this time, the interaction intensity processing model can be used to predict the interaction intensity between the target account and the second candidate publishing account based on the account feature data of the target account and the account feature data of the second candidate publishing account, so as to obtain the account interaction intensity between the target account and the second candidate publishing account.

[0057] In this case, the second candidate publishing account with an interaction intensity greater than the interaction intensity threshold can be regarded as at least one content publishing account that has an interaction relationship with the target account, or the specified number of second candidate publishing accounts with the highest interaction intensity can be determined as at least one content publishing account.

[0058] The target account's account characteristic data can include basic account information and interaction data. Basic account information can include the target account's age, gender, region, account tags, and identity verification. Interaction data can include the account identifier of the followed account, the basic account information of the followed account, and content description data of the content posted. Content description data can include content topics carried in the posted content, content tags obtained from content analysis (e.g., low-fat afternoon tea), and content scenarios (e.g., fitness scenarios). No specific limitations are imposed here.

[0059] The account characteristic data for the second candidate posting account can include basic account information and posting data. Basic account information can include age, gender, region, account tags, identity verification, number of followers, etc. Posting data can be the content description data of the most recently posted content, or the content description data of a specified number of posts with the highest number of likes, etc.

[0060] Specifically, the account feature data of the target account and the account feature data of the second candidate posting account can be input into the interaction intensity processing model. The interaction intensity processing model performs feature fusion on the account feature vector of the target account obtained from the account feature data of the target account and the account feature vector of the second candidate posting account obtained from the account feature data of the second candidate posting account, to obtain a cross-account feature vector for the target account and the second candidate posting account. The account interaction intensity is then predicted based on the cross-account feature vector.

[0061] Therefore, for content-posting accounts with interactive relationships, not only can the interaction strength between the target account and candidate posting accounts be statistically analyzed, but also online predictions can be performed using neural network models to fill in sparse or recent behavioral relationships. Subsequently, a unified affinity service interface can be used to provide a unified affinity query capability (Affinity API), thereby obtaining the content-posting accounts of the target account.

[0062] The account selection strategy for selecting the second candidate posting account may include at least one of the following: posting accounts that have interacted with the target account during the historical period, posting accounts associated with posting accounts that have interacted with the target account during the historical period, accounts whose pages have been viewed by the target account during the historical period, and accounts that are in the same group chat as the target account and are active at the same time as the target account during the historical period.

[0063] The second candidate posting account refers to an account with a potential connection, that is, an account that may be of interest to the target account. Therefore, any account that has had interaction or social relationship can be used as a second candidate posting account.

[0064] For example, accounts that the target account has liked or favorited during a specific time period. Alternatively, accounts associated with accounts that the target account has interacted with (such as accounts followed by the target account, accounts that are friends of the target account, or accounts that interact frequently with the target account). Accounts whose pages the target account has viewed. Alternatively, accounts in the same group chat as the target account with high group chat activity (i.e., accounts whose activity is determined by their posts). There are no restrictions on the account selection strategy here.

[0065] Optionally, if the content publishing account is obtained from the first candidate publishing account, a black-box screening strategy can also be used to select the content publishing account from the first candidate publishing account. Alternatively, if the content publishing account is obtained from the second candidate publishing account, a white-box screening strategy can also be used to select the content publishing account from the second candidate publishing account.

[0066] Alternatively, if the content publishing account is obtained from the first and second candidate publishing accounts, a black-box screening strategy and a white-box screening strategy can be used respectively, or both can be selected using one strategy. Optionally, if duplicate publishing accounts exist between the first and second candidate publishing accounts, deduplication can be performed, for example, removing accounts that appear in the first candidate publishing account list from the second candidate publishing account list. No specific selection strategy for content publishing accounts is limited here.

[0067] When training the interaction intensity processing model, the interaction intensity of the sample accounts used to train the model can be determined by a white-box screening strategy, that is, by determining the interaction intensity of the sample accounts through statistical interaction data.

[0068] For example, you can obtain the first sample account used to train the initial interaction intensity processing model, and the second sample account that has interacted with the first sample account (you can obtain accounts that have interacted with each other, or accounts that have potential relationships).

[0069] Specifically, this involves obtaining interaction data from the first sample account across N social scenarios within a historical timeframe, targeting the second sample account. Based on this interaction data across the N social scenarios, the intensity of interaction between the first and second sample accounts in each scenario can be determined. This allows for the determination of the overall interaction intensity between the first and second sample accounts across these N social scenarios.

[0070] At this point, the initial interaction intensity processing model can be trained based on the interaction intensity of the sample accounts to obtain the trained interaction intensity processing model. Specifically, this can be achieved by using the initial interaction intensity processing model to predict the interaction intensity between the first and second sample accounts based on the account feature data of the first and second sample accounts, thus obtaining the predicted account interaction intensity between the first and second sample accounts; based on the sample account interaction intensity and the predicted account interaction intensity, the initial interaction intensity processing model can be trained to obtain the trained interaction intensity processing model.

[0071] In other words, the interaction intensity processing model in the black-box screening strategy can be trained using the account interaction intensity determined by the white-box screening strategy.

[0072] Once at least one content publishing account is identified, the content to be recommended to the target account can be retrieved from that account. For example, the first piece of content from each account that meets a quality threshold can be retrieved. The content quality can be determined using a content quality model (e.g., inputting content data, number of favorites, number of comments, number of likes, etc., into the model to predict the corresponding content quality). Alternatively, the content quality can be obtained by weighted summation of interaction data (number of favorites, number of comments, number of likes).

[0073] This can be achieved by selecting high-quality content based on a quality threshold, or by prioritizing a specified number of high-quality posts as the first set of posts. Therefore, a subset of high-quality posts can be selected from each content posting account to form the first candidate content set.

[0074] Optionally, for the first published content selected from the content published by each content publishing account, quantity limits and category deduplication controls can be implemented to prevent excessive concentration of content from a single content publishing account or a single theme. For example, there can be a limit to the number of first published content items obtained for each content publishing account (e.g., a maximum of 10). In addition, if there are similar themes among the first published content items, the content with the highest quality can be retained.

[0075] For example, the content similarity between any two first published content can be determined (e.g., determining content similarity based on the content data of the published content). If it is determined that there are two similar first published content based on content similarity (e.g., content similarity greater than 90%), then only the published content with higher content quality can be retained.

[0076] S102. Obtain at least one similar publishing account that has similar publishing attributes to each content publishing account. Based on the account exposure of the similar publishing account, obtain the target similar publishing account from the at least one similar publishing account. Obtain the second publishing content with the content quality reaching the quality threshold from the publishing content published by the target similar publishing account, as the second candidate content set.

[0077] To enrich the list of accounts recommended to the target account, it's also possible to identify more accounts that users might be interested in based on the content-publishing account. For example, similar accounts can be identified, such as accounts in the same field (e.g., accounts with the same verified identity as the content-publishing account, such as both being travel bloggers, or accounts whose content is similar in theme to the content-publishing account, such as both posting cooking-related content). In other words, accounts with similar posting attributes to the content-publishing account can be identified as similar posting accounts. When the target account interacts with the content-publishing account, these similar posting accounts may become accounts that the target account is also interested in, and therefore, their content can be recommended to the target account.

[0078] For example, at least one content publishing account includes the target publishing account. A specific method for obtaining at least one similar publishing account could be: selecting candidate similar accounts from the account set that have associated account tags with the target publishing account; determining the publishing attribute data of the target publishing account based on its interaction data and published content, and determining the publishing attribute data of the candidate similar accounts based on their interaction data and published content; using an attribute similarity processing model to predict the attribute similarity between the target publishing account and the candidate similar accounts based on their publishing attribute data, thus obtaining the publishing attribute similarity between the target publishing account and the candidate similar accounts; and identifying candidate similar accounts with publishing attribute similarity greater than a similarity threshold as at least one similar publishing account that shares similar publishing attributes with the target publishing account.

[0079] Among them, the account tag can be the identity verification of the target posting account, or the tag obtained by analyzing the content scenario represented by the posted content (for example, if you frequently post content related to food exploration, the tag obtained by analysis can be the food exploration tag; if you frequently post content related to fitness, the tag obtained by analysis can be the fitness tag).

[0080] The "account set" refers to accounts on a content platform that have published content. The candidate similar accounts are those that may have a similar relationship to the target publishing account.

[0081] Among them, related account tags can refer to identical account tags or similar account tags (for example, similarity can be determined based on the field to which the account tags belong, such as the food exploration tag and the cooking tutorial tag belonging to the food field, so they can be considered similar account tags; similarly, the fitness tag and the sports equipment recommendation tag belonging to the sports field can be considered similar account tags. Alternatively, similarity can be determined based on the tag similarity, such as account tag 1 being a food exploration tag and account tag 2 being a food mukbang tag, account tag 1 and account tag 2 can be determined to be similar account tags based on tag similarity).

[0082] The target posting account and candidate similar accounts can refer to accounts with similar interests, accounts whose followers share similarities, or accounts that post similar content. Therefore, the posting attribute data of the target posting account can be obtained, such as interaction data (e.g., account identifiers or tags of followed accounts, account tags or tags of follower accounts, content identifiers of saved posts, content identifiers of liked posts, etc.). This can also include the posted content itself, such as the content data (or content tags) of the most recently posted content, the content data of the posts with the most likes or saves, and the type of live stream (e.g., chat live stream, shopping live stream, etc.). No further limitations are imposed here.

[0083] Specifically, the posting attribute data of the target posting account and the posting attribute data of candidate similar accounts can be input into the attribute similarity processing model to obtain the posting attribute similarity between the target posting account and the candidate similar accounts. For example, the posting attribute feature vector corresponding to the posting attribute data can be determined, and the posting attribute similarity can be determined based on the vector similarity between the posting attribute feature vector of the target posting account and the posting attribute feature vector of the candidate similar accounts.

[0084] Among them, candidate similar accounts with a similarity of posting attributes greater than a similarity threshold can be selected as similar posting accounts, or a specified number of candidate similar accounts with the highest similarity threshold can be selected as similar posting accounts.

[0085] For each content publishing account, a subset of similar publishing accounts can be obtained. Alternatively, a second post can be selected directly based on these similar accounts, or a further filter can be applied to select target similar posts, and then a second post can be chosen based on these target similar posts.

[0086] Among them, similar posting accounts can be selected based on account exposure. That is, for top authors on content platforms, the more attention they receive, the more likely they are to be accounts that the public is interested in. Therefore, accounts with higher exposure can be selected from similar posting accounts as target similar posting accounts.

[0087] In this case, determining account exposure can be based on the interactive actions of other accounts towards accounts that post similar content (such as determining account exposure through interactive actions such as following, browsing the homepage, viewing the posted content, and watching live streams).

[0088] For example, it could be based on one or more of the following factors, the number of interactions with the published content, and the number of views of each similar posting account in a historical period to determine the account exposure of at least one similar posting account in a historical period.

[0089] For example, you can obtain the current number of followers of similar accounts (or the number of new followers added during a historical period). Alternatively, you can obtain the number of interactions with the published content (e.g., the weighted sum of likes, favorites, and comments for all published content; or the weighted sum of likes, favorites, and comments for content published during a historical period). Alternatively, you can obtain the current number of page views (i.e., the number of views on the account's homepage) for the similar account, or the number of page views added during a historical period. Alternatively, you can obtain the number of interactions with live streams published by similar accounts (e.g., likes, sales volume, number of viewers, etc.), or the number of interactions with live streams published during a historical period. Here, there are no restrictions on the account exposure data used to determine account exposure.

[0090] This involves weighting and summing one or more of the following data points: number of followers, number of interactions with the published content, number of views, etc., to obtain the account exposure of similar accounts within a historical period.

[0091] Specifically, accounts with an exposure volume greater than a threshold among at least one similar posting account can be identified as target similar posting accounts. Accounts with an exposure volume greater than the threshold can be considered top authors within a historical period (e.g., top authors within the current month; alternatively, exposure volume from other historical periods, such as top authors within the past three months).

[0092] There can be one or more accounts with similar targets, and high-quality content can be extracted from the posts of each account as the second set of candidate content. The specific method for extracting the second set of candidate content can be the same as the method for extracting the first set of candidate content.

[0093] S103. When constructing the target content set through the first candidate content set and the second candidate content set, recommend the published content in the target content set to the target account.

[0094] The process involves using the first and second candidate content sets as the recommended content selected through various filtering strategies to obtain the target content set. Content from this target content set can then be recommended to the target account. For example, the content in the target content set can be sorted, and the sorted content can be recommended to the target account sequentially. Furthermore, the content in the target content set can be finely scored, such as by predicting click-through rate (CTR) and conversion rate (CVR), and then sorted by CTR and CVR to obtain the content that best matches user interests.

[0095] This could involve recommending all published content from the target content set to the target account sequentially, or selecting a portion of published content from the target content set and recommending it to the target account.

[0096] The process of selecting published content from the target content set can be understood as a preliminary ranking to obtain recommendable content. Subsequent fine-tuning can be performed on the published content in the target content set to select content recommended to the target account. For example, this could involve determining the interaction rate of the target account for each piece of content in the target content set (e.g., interaction rate obtained through click-through rate and / or conversion rate). The target content set can then be sorted according to the interaction rate, resulting in a sorted target content set. Recommended content for the target account can then be selected from this sorted set and recommended to the target account.

[0097] For example, the first specified number of published content items in the sorted target content set can be used as recommended published content, or the published content items in the sorted target content set with an interaction rate greater than an interaction rate threshold can be used as recommended published content.

[0098] For example, such as Figures 4-5 As shown, Figures 4-5 This is a schematic diagram illustrating a content recommendation scenario provided in an embodiment of this application; wherein, in Figure 4In this process, a white-box screening strategy can be used to obtain the first candidate posting account that has interacted with the target account. The interaction data of the target account against the first candidate posting account in N social scenarios (such as social scenario 1 and social scenario 2) within a historical period can be counted, and the intensity of the scenario interaction between the target account and the first candidate posting account in each social scenario can be determined.

[0099] Specifically, the interaction strength between the target account and the first candidate publishing account can be determined based on the interaction strength in N social scenarios, and the content publishing account can be determined from the first candidate publishing accounts based on the interaction strength.

[0100] Alternatively, a black-box screening strategy can be used to obtain a second candidate posting account that has a potential relationship with the target account. The account feature data of the target account and the second candidate posting account can be input into the interaction intensity processing model to obtain the account interaction intensity between the target account and the second candidate posting account.

[0101] Among these, the content publishing account can be determined from the second candidate publishing accounts based on the account's interaction intensity.

[0102] Among these, the first published content that meets the quality threshold can be selected from the content published by the content publishing account as the first candidate content set.

[0103] Among them, Figure 5 In this process, the account tags of the target posting account can be determined, and candidate similar accounts with associated account tags with the target posting account can be selected from the account set. The posting attribute data of the target posting account and the posting attribute data of the candidate similar accounts are input into the attribute similarity processing model to obtain the posting attribute similarity between the target posting account and the candidate similar accounts. Based on the posting attribute similarity, similar posting accounts with similar posting attributes to the target posting account can be obtained from the candidate similar accounts.

[0104] This involves obtaining the number of followers, interactions, and views of similar accounts within a historical period to determine their exposure. Based on this exposure, target similar accounts are identified from among the similar accounts.

[0105] Among these, the second set of candidate content can be selected from the content published by accounts with similar targets, whose content quality reaches the quality threshold.

[0106] Specifically, a target content set can be constructed based on the first candidate content set and the second candidate content set to realize the content to be published in the target content set.

[0107] Therefore, at the front end of the recommendation chain, content recommendations can be made to returning users and new users based on the "closeness" and "similarity" between users and authors. This aims to activate dormant users and increase clicks, interactions, engagement, and long-term retention of new users. Specifically, this can be achieved by strengthening the exposure of content from authors with high historical closeness and similar authors, thereby rekindling the user's memory and trust in the platform, promoting their reactivation, and recalling high-quality content with long-term value for both returning and new users.

[0108] Optionally, in addition to recommending published content, you can also recommend products, live streams, etc. There are no restrictions here.

[0109] It's understandable that in content communities and news platforms, the "intimacy" between users and authors is a key indicator for measuring the quality of long-term relationships. Compared to focusing solely on single clicks or short-term behaviors, intimacy prioritizes conversion rates after multiple days, scenarios, and types of engagement, thus offering a natural advantage for reactivating returning users and initial recommendations for new users. For instance, for users returning after uninstalling, these users have historically engaged in continuous and repeated engagement and conversion behaviors with certain authors over a period of time (e.g., browsing, interacting, sending private messages, watching live streams, etc.). Therefore, the intensity of account interaction between the target account and these authors can be determined based on the combined signal of "multi-day engagement + effective conversion." Consequently, when users return, intimacy can quickly identify "strong relationship authors" they previously trusted and whose relationships have been verified over multiple days. Using these authors and their content as entry points for recall and recommendation can effectively rekindle user memory and trust, significantly improving the success rate of return activation and subsequent long-term activity.

[0110] For example, for new users, after they have generated a small amount of initial behavior (such as following several authors, repeatedly browsing or lightly interacting with some authors within a few days), intimacy can be amplified and stabilized within the framework of "multi-day reach and conversion," quickly identifying the set of authors "with the potential to evolve into long-term strong relationships." Using these highly intimate authors as the initial core of recommendations for new users is not only closer to real interests than simple daily clicks or popular recommendations, but also helps to establish a solid relationship network early in the user lifecycle, providing long-term support for subsequent interaction rates and retention.

[0111] Therefore, the content recommendation method constructed in this application, based on the premise of account interaction intensity, has significant recommendation value in two key scenarios: reactivating returning users and initial recommendation of new users, from a long-term perspective of multi-day reach and conversion. It can both build on past relationship accumulation and provide a stable and reliable measurement basis for cultivating new relationships in the early stage.

[0112] In this embodiment, at least one content publishing account that has an interactive relationship with the target account during a historical period can be obtained. That is, by account, content publishing accounts that are closely related to the target account can be obtained based on the target account's behavior. This indicates that the target account is more likely to be interested in the content published by such accounts. Therefore, the first published content that meets the quality threshold can be obtained from the published content of each content publishing account, i.e., high-quality content published by authors with high affinity to the target account is selected as the first candidate content set. Furthermore, to enrich the source of the selected content, at least one similar publishing account with similar publishing attributes to each content publishing account can also be obtained. The published content published by authors similar to the high-affinity authors may also be published content that users are interested in. Therefore, similar publishing accounts can be used as a basis for further selection. The method involves obtaining target similar publishing accounts from at least one similar publishing account, and then extracting second-tier publishing content that meets a quality threshold from the content published by the target similar publishing accounts. This second-tier content set can be used to filter out content published by similar authors. The target content set can then be constructed using the first and second-tier content sets to achieve content recommendation. Compared to directly determining a user's interest in each published content from the published content library, the proposed method allows for content filtering from richer dimensions, such as filtering content to be recommended based on highly intimate exclusive accounts or similar accounts. This makes it more likely to identify published content that users are interested in, thereby increasing the diversity of content recommendation methods and improving the richness and effectiveness of recommended content.

[0113] Based on the above description, this application proposes a content recommendation method, which can be executed by the aforementioned electronic device, specifically... Figure 1 The server shown. Please refer to [link / reference]. Figure 6 , Figure 6 A flowchart illustrating a content recommendation method provided in this application embodiment. Figure 2 .like Figure 6 As shown, the content recommendation method of this application embodiment may include the following flow: S201. Obtain at least one content publishing account that has an interactive relationship with the target account during the historical period, and obtain the first published content whose content quality reaches the quality threshold from the published content of each content publishing account, as the first candidate content set.

[0114] S202. Obtain at least one similar publishing account that has similar publishing attributes to each content publishing account. Based on the account exposure of the similar publishing accounts, obtain a target similar publishing account from the at least one similar publishing account. Then, obtain second publishing content whose content quality reaches a quality threshold from the publishing content published by the target similar publishing account, as a second candidate content set. The specific implementation methods of steps S201-S202 can be found in the relevant descriptions of the above embodiments, and will not be repeated here.

[0115] S203. When a third candidate content set is obtained by recalling the published content set according to the content recall strategy, the first candidate content set, the second candidate content set, and the third candidate content set are determined as the target content set.

[0116] Among them, some published content can be recalled according to other basic recall methods (such as content recall strategies such as recalling some pushed content, recalling some highly popular content, recalling other published content published by the account that published the content that the user clicked on during the last refresh, etc.). In this case, the first candidate content set, the second candidate content set, and the third candidate content set can be used as the target content set obtained by recall.

[0117] Therefore, compared to relying solely on other basic recall methods to recall published content for content recommendation, this application can enrich the recall sources through two filtering strategies (filtering by content publishing account and filtering by similar publishing accounts). At the same time, it can filter out multiple authors that users may be interested in based on the degree of interaction and relevance between the target account and the author, thereby obtaining the content published by that author for content recommendation and thus improving the content recommendation effect.

[0118] S204. Determine the content interaction rate of each published content in the target content set for the target account, and sort the target content set according to the content interaction rate to obtain the sorted target content set.

[0119] The target content set can be sorted based on its content interaction rate. This interaction rate can be obtained from one or more of the following: estimated click-through rate, estimated completion rate (e.g., video completion rate if the content is video), and estimated interaction probability (e.g., probability of liking, commenting, or saving). A weighted summation can be used to obtain the content interaction rate. Alternatively, a neural network model can be used to predict the content interaction rate based on relevant account data of the target account and relevant content data of the published content.

[0120] S205. Select recommended content for the target account from the sorted target content set and recommend the recommended content to the target account.

[0121] Among them, the recommended published content includes at least M target published content, and each target published content comes from the first candidate content set or the second candidate content set, where M is a positive integer.

[0122] This can be achieved by selecting the top specified number of published content items from the sorted target content set as recommended content, or by selecting published content items with an interaction rate greater than a threshold as recommended content.

[0123] Optionally, a robust protection mechanism can be provided for the first or second candidate content set to ensure that the recommended published content includes at least M target published content items. These target published content items refer to published content items from either the first or second candidate content set.

[0124] For example, if the number of target content items in the determined recommended content is less than M, then target content items are sequentially obtained from the sorted target content set as recommended content items to supplement the recommended content.

[0125] For example, if the top 30 published content items in the sorted target content set are selected as recommended published content, and there are 5 target published content items among these top 30 published content items, and M is 10, then starting from the 31st published content item in the sorted target content set, the target published content items need to be selected sequentially (sequentially checking whether the published content item belongs to the first candidate content set or the second candidate content set; if it belongs to the first candidate content set or the second candidate content set, it is selected as the target published content item and added to the recommended published content), until the number of target published content items in the recommended published content set is greater than or equal to M.

[0126] Alternatively, one could select some published content from the first, second, and third candidate content sets as recommended content. This could be done by sorting and filtering them separately.

[0127] The specific method for selecting recommended published content from the target content set is not limited. The purpose of this application is to enrich the content selection source of the recommended published content and ensure that the recommended published content selected from the target content set includes at least M published content from the first candidate content set and the second candidate content set (i.e., the published content determined by the technical solution of this application).

[0128] Optionally, the recommended content can be further refined and ranked to sequentially recommend it to target accounts. This ranking could be based on factors such as diversity, freshness, and business rules to avoid clustering of similar or related content, appropriately insert new content, explore potential user interests (e.g., prioritizing content with low engagement rates), and meet operational needs, such as ensuring exposure for important content (e.g., prioritizing content pushed to the front of the feed). Specific strategies are not limited here.

[0129] For example, such as Figure 7 As shown, Figure 7 This illustration shows a content recommendation scenario provided by an embodiment of this application. The technical solution of this application proposes a content recommendation method primarily targeting users returning from uninstallation and new users. This method employs a hybrid dual-strategy content recall approach based on "account interaction strength, similar accounts, and a strong protection mechanism." This method constructs a high-quality candidate content set at the front end of the recommendation chain and ensures stable exposure in the ranking chain, thereby improving the reactivation effect of returning users and the long-term retention of new users.

[0130] The specific steps are as follows: Step 1, the intimate account acquisition module. This module is used to construct rules for acquiring content publishing accounts, such as statistical calculation rules under a white-box screening strategy and interaction intensity processing models under a black-box screening strategy. It also deploys a module interface for calling the intimate account acquisition module. Specifically, it calculates the interaction intensity between the target account and the author to identify high-intensity accounts.

[0131] Specifically, when it is necessary to recommend content to a target account, the module interface can be called to obtain the content publishing account from the intimate account acquisition module (which can be calculated in real time or at fixed times, such as once a day for the target account), such as based on statistical calculation rules or based on the interaction intensity processing model.

[0132] In step two, the first candidate content set construction module can obtain the first published content from the published content of the content publishing account when obtaining the content publishing account through the intimate account acquisition module, thus obtaining the first candidate content set.

[0133] Therefore, account interaction intensity can be leveraged to build a high-value candidate pool that is more conducive to long-term retention for both returning and new users. Specifically: First, identify the account type of the target account. When targeting returning and new users, obtain the first candidate publishing accounts through the close account acquisition module and calculate the account interaction intensity to filter out content publishing accounts, i.e., the "high-closeness author set." This prioritizes retaining authors with more stable relationships and verified engagement over several days. Then, introduce quality thresholds (such as likes, total interactions, and completion rate) to filter high-quality published content, ensuring that the content recommended to new or returning users is both familiar and consistently attractive. Optionally, if necessary, limit the number of high-quality published content and control category deduplication to avoid excessive concentration by a single author or theme. Finally, generate a high-value candidate pool for target users consisting of high-quality representative works from high-closeness authors. This allows subsequent collaborative recall and coarse-grained selection modules to prioritize distributing content more conducive to activation and long-term retention when targeting returning and new users.

[0134] In step three, the second candidate content set construction module can obtain similar publishing accounts to the content publishing account to obtain the second published content and thus the second candidate content set. Optionally, in addition to obtaining similar publishing accounts to the content publishing account, it can also obtain similar publishing accounts to accounts that have positively commented on the target account (such as similar accounts to accounts that have recently performed interactive operations, similar accounts to accounts that published content that was clicked on during the last refresh, or similar accounts to live streaming accounts that viewed the live stream within a specified time, such as within 10 minutes, etc., which are not limited here).

[0135] After completing the construction of the first candidate content set, similar accounts can be further introduced to expand the candidate space and align with users' historical preferences, thereby constructing the second candidate content set.

[0136] For example, firstly, a set of users' recent positive author history is obtained based on their recent positive behaviors (such as clicks, completions, and interactions), and this set is merged with a set of highly relevant authors to form an anchor author set. Then, for each anchor author, similar authors are searched to form a candidate set of similar authors (i.e., similar publishing accounts). To ensure the freshness and quality of the published content, the intersection of this candidate set of similar authors and the top-performing authors of the past few days is taken (i.e., selecting target similar publishing accounts based on account exposure), retaining only the published content from authors with recent high performance and similarity to the anchor authors. In this way, while maintaining users' familiarity with authors with strong relationships, high-quality new authors and content similar to them are introduced, resulting in a combination of "familiarity and freshness," thus improving clicks, interactions, and interest expansion.

[0137] In step four, the coarse sorting and strict protection module, the target content set obtained from the first, second, and third candidate content sets is sorted to obtain a sorted target content set. Recommended content is then selected from this sorted target content set. Optionally, during the selection of recommended content, a strict protection mechanism can be used to ensure that a certain number of recommended posts originate from the first and second candidate content sets.

[0138] After obtaining the first candidate content set based on affinity and the second candidate content set from author-coordinated recall, coarse ranking and strict protection controls can be applied to the multi-path recall results. For example, the first candidate content set, the second candidate content set, and the third candidate content set obtained from other basic recall sources can be merged, and the overall candidate set (target content set) for each user can be input into the coarse ranking model. Based on the estimated click-through rate, estimated completion rate, and estimated interaction probability, a coarse ranking score can be generated and sorted.

[0139] Subsequently, a strong protection mechanism was introduced: for candidates whose sources are marked as "first candidate content set" or "second candidate content set", at least a certain number of published content will be reserved for the target account (for example, a certain number of published content can be reserved for both "first candidate content set" and "second candidate content set", or a certain number of published content can be reserved for each of "first candidate content set" and "second candidate content set"). These will enter the fine ranking process, and even if their rough ranking score is slightly lower, they will not be eliminated as a whole. This avoids key relationship authors and high-potential content from being completely eliminated due to a single low score, ensuring that such content has stable exposure in the final recommendation and improving the stability of relationships with returning users and new users.

[0140] Optionally, unified content security checks and filtering of unhealthy content will still be performed. This mechanism ensures stable exposure of the "first candidate content set" or "second candidate content set" in the final recommendation, which is beneficial for improving user reactivation and new user retention.

[0141] In step five, within a recommendation scenario, the recommended content can be re-ranked using fine-tuning, and then presented sequentially to the target account. This target account can be a returning user who has uninstalled the app, or a new user.

[0142] For training the interaction intensity processing model, users who uninstalled and returned, as well as new users, can be used as the first sample accounts, and corresponding second sample accounts can be obtained. The interaction intensity between the first and second sample accounts can be obtained according to the statistical calculation rules mentioned above. At this time, the account feature data of the first and second sample accounts can be obtained to train the interaction intensity processing model.

[0143] In this application, the intimate account acquisition module, the first candidate content set construction module, the second candidate content set construction module, and the coarse ranking and strong protection modules are uniformly deployed as online services, embedded in the front end of the recommendation link, and respond to recommendation requests from uninstalled and returning users and new users in real time.

[0144] Optionally, feedback data such as online exposure, clicks, reads, interactions, follows, returnee activation, and long-term retention can be collected periodically for continuous optimization. This includes updating model parameters for interaction intensity processing models, weighting coefficients in statistical calculation rules, and adjusting relevant thresholds. This enables continuous adaptive optimization in response to changes in user behavior and business goals. It effectively rekindles the memory and trust of returning users while reducing low-quality content entering the ranking process, thus improving overall recommendation performance. This allows for the construction of an online service and closed-loop optimization system around "account interaction intensity, similar accounts, and strong protection mechanisms." This comprehensive solution is used for long-term tracking and optimization of the recall effect of uninstalled returning users and new users, giving the entire recall strategy self-learning and continuous evolution capabilities. It can automatically adjust to changes in user behavior and business goals, maintain good results, and is easily migrated to other recommendation scenarios (such as live streaming recommendations and product recommendations).

[0145] In this embodiment, at least one content publishing account that has an interactive relationship with the target account during a historical period can be obtained. Specifically, content publishing accounts closely associated with the target account can be identified based on the target account's behavior. From the content published by each content publishing account, the first published content meeting a quality threshold is selected; that is, high-quality content published by authors with high affinity to the target account is selected as the first candidate content set. Furthermore, to enrich the source of the selected content, at least one similar publishing account with similar publishing attributes to each content publishing account can be obtained. The target similar publishing account can be obtained from at least one similar publishing account based on the account exposure of the similar publishing accounts. The method involves identifying second-tier content that meets a quality threshold from similar accounts and using this second-tier content set as a secondary candidate set. Furthermore, other content recall strategies can be used to select more third-tier candidate content sets, thus constructing a target content set for content recommendation. Compared to directly determining user interest in each post from the content library, this method allows for content filtering from richer dimensions, such as filtering from highly relevant or similar accounts. This makes it more likely to identify posts that users are interested in, thereby increasing the diversity of content recommendation methods and improving the richness and effectiveness of the recommended content.

[0146] Please see Figure 8 , Figure 8This is a schematic diagram of a content recommendation device provided in an embodiment of this application. It should be noted that... Figure 8 The content recommendation device shown is used to perform the present application. Figure 3 , Figure 6 The methods in the illustrated embodiments are shown only in the parts relevant to the embodiments of this application for ease of explanation; specific technical details are not disclosed. Reference to this application is required. Figure 3 , Figure 6 The embodiment shown. The content recommendation device 900 may include: a content acquisition module 901 and a content recommendation module 902. Wherein: The content acquisition module 901 is used to acquire at least one content publishing account that has an interactive relationship with the target account during the historical period, and to acquire the first published content whose content quality reaches the quality threshold from the published content of each content publishing account, as the first candidate content set; The content acquisition module 901 is also used to acquire at least one similar publishing account that has similar publishing attributes to each content publishing account, acquire a target similar publishing account from the at least one similar publishing account based on the account exposure of the similar publishing account, and acquire second publishing content with content quality reaching the quality threshold from the publishing content published by the target similar publishing account as a second candidate content set; The content recommendation module 902 is used to recommend published content from the target content set to the target account when the target content set is constructed by the first candidate content set and the second candidate content set.

[0147] Specifically, when the content acquisition module 901 acquires at least one content publishing account that has an interactive relationship with the target account within a historical period, it is used for: Obtain the first candidate posting accounts that have interacted with the target account; account interactions include at least one of the following: following behavior, chat behavior; Obtain the interaction data of the target account against the first candidate posting account in N social scenarios within a historical time period, and determine the scene interaction intensity between the target account and the first candidate posting account in each social scenario based on the interaction data in N social scenarios; N is a positive integer; The interaction strength between the target account and the first candidate publishing account is determined based on the interaction strength in N social scenarios. The first candidate publishing account with an interaction strength greater than the interaction strength threshold is regarded as at least one content publishing account that has an interaction relationship with the target account.

[0148] Specifically, when the content acquisition module 901 acquires at least one content publishing account that has an interactive relationship with the target account within a historical period, it is used for: Based on the account selection strategy, a second candidate posting account with a potential relationship to the target account is selected from the account set; The interaction intensity processing model predicts the interaction intensity between the target account and the second candidate posting account based on the account feature data of the target account and the account feature data of the second candidate posting account, thus obtaining the account interaction intensity between the target account and the second candidate posting account. The second candidate publishing account with an interaction intensity greater than the interaction intensity threshold is considered as at least one content publishing account that has an interaction relationship with the target account; The account selection strategy includes at least one of the following: accounts that have posted content that the target account has interacted with during the historical period; accounts associated with accounts that have posted content that the target account has interacted with during the historical period; accounts whose pages have been viewed by the target account during the historical period; and accounts that are in the same group chat as the target account and whose group chat activity level is greater than the preset activity level during the historical period.

[0149] The content acquisition module 901 is also used for: Obtain the first sample account used to train the initial interaction intensity processing model, and obtain the second sample account that has interacted with the first sample account; Obtain the interaction data of the first sample account against the second sample account in N social scenarios within a historical period, and determine the scene interaction intensity between the first sample account and the second sample account in each social scenario based on the interaction data in N social scenarios; N is a positive integer; The interaction intensity between the first and second sample accounts is determined based on the interaction intensity in N social scenarios. The initial interaction intensity processing model is trained based on the interaction intensity of sample accounts to obtain the trained interaction intensity processing model.

[0150] Specifically, when the content acquisition module 901 is used to train the initial interaction intensity processing model based on the interaction intensity of sample accounts to obtain the trained interaction intensity processing model, it is used for: The initial interaction intensity processing model predicts the interaction intensity between the first sample account and the second sample account based on the account feature data of the first sample account and the account feature data of the second sample account, thus obtaining the predicted account interaction intensity between the first sample account and the second sample account. Based on the interaction strength of sample accounts and the interaction strength of predicted accounts, the initial interaction strength processing model is trained to obtain the trained interaction strength processing model.

[0151] At least one of the content publishing accounts includes the target publishing account; When the content acquisition module 901 is used to acquire at least one similar publishing account that has similar publishing attributes to each content publishing account, it is specifically used for: Select candidate similar accounts from the account set that have related account tags to the target posting account; The posting attribute data of the target posting account is determined based on the interaction data and the posting content of the target posting account, and the posting attribute data of the candidate similar accounts is determined based on the interaction data and the posting content of the candidate similar accounts. The attribute similarity processing model predicts the attribute similarity between the target posting account and the candidate similar accounts based on the posting attribute data of the target posting account and the posting attribute data of the candidate similar accounts, thus obtaining the posting attribute similarity between the target posting account and the candidate similar accounts. Candidate similar accounts with a similarity score greater than the similarity threshold are considered as at least one similar posting account that has similar posting attributes to the target posting account.

[0152] Specifically, when the content acquisition module 901 is used to obtain a target similar posting account from at least one similar posting account based on the account exposure of similar posting accounts, it is used for: Based on one or more of the following factors, the number of interactions with the published content, and the number of views of each similar posting account in the historical period, determine the account exposure of at least one similar posting account in the historical period. Accounts with an exposure volume greater than the exposure threshold among at least one similar posting account are identified as target similar posting accounts.

[0153] Specifically, when the content recommendation module 902 recommends published content from the target content set to the target account, it is used for: Determine the content interaction rate of the target account for each piece of content published within the target content set; The target content set is sorted according to the content interaction rate to obtain the sorted target content set; Select recommended content for the target account from the sorted target content set, and recommend the recommended content to the target account.

[0154] Specifically, when the content recommendation module 902 recommends published content from the target content set to the target account after constructing the target content set using the first and second candidate content sets, it is used for: When the third candidate content set is obtained by recalling the published content set according to the content recall strategy, the first candidate content set, the second candidate content set, and the third candidate content set are determined as the target content set; Determine the content interaction rate of the target account for each piece of content published within the target content set; The target content set is sorted according to the content interaction rate to obtain the sorted target content set; Select recommended content for the target account from the sorted target content set and recommend the recommended content to the target account; the recommended content includes at least M target content, and each target content comes from the first candidate content set or the second candidate content set, where M is a positive integer.

[0155] The target account refers to at least one of the following: an account that registers the target application for the first time within a preset time period, or an account that downloads and logs into the target application on a terminal device within a preset time period, then uninstalls the target application and re-downloads and logs into the target application.

[0156] The specific implementation methods of the content acquisition module and the content recommendation module can be found in the description of the above embodiments, and will not be repeated here. It should be understood that the beneficial effects obtained by using the same method will also not be repeated here.

[0157] In the embodiments of this application, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0158] Please see Figure 9 , Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 9As shown, the electronic device 2600 includes at least one processor 2601 and a memory 2602. Optionally, the electronic device may also include a network interface. The processor 2601, memory 2602, and network interface can exchange data. The network interface, controlled by the processor 2601, is used to send and receive messages. The memory 2602 stores computer programs, including program instructions. The processor 2601 executes the program instructions stored in the memory 2602. The processor 2601 is configured to invoke the program instructions to execute the aforementioned method. The memory 2602 may include volatile memory, such as random-access memory (RAM); the memory 2602 may also include non-volatile memory, such as flash memory, solid-state drive (SSD), etc.; the memory 2602 may also include combinations of the above types of memory.

[0159] Processor 2601 may be a central processing unit (CPU). In one embodiment, processor 2601 may also be a graphics processing unit (GPU). Processor 2601 may also be a combination of a CPU and a GPU. Processor 2601 may be used to invoke device control applications stored in memory 2602 to perform the above-described tasks. Figure 3 , Figure 6 The description of the content recommendation method in the corresponding embodiments can also be executed as described above. Figure 8 The description of the content recommendation device in the corresponding embodiments will not be repeated here. Furthermore, the beneficial effects of using the same method will also not be repeated.

[0160] In specific implementations, the devices, processors, memory, etc., described in the embodiments of this application can execute the implementation methods described in the above method embodiments, or they can execute the implementation methods described in the embodiments of this application, which will not be repeated here.

[0161] This application also provides a computer-readable storage medium storing a computer program. The computer program includes program instructions, which, when executed by a processor, enable the processor to perform some or all of the steps described in the above method embodiments. Optionally, the computer storage medium can be volatile or non-volatile. The computer-readable storage medium may primarily include a program storage area and a data storage area. The program storage area may store an operating system, at least one application program required for a given function, etc.; the data storage area may store data created based on the use of blockchain nodes, etc.

[0162] This application provides a computer program product, which may include a computer program. When the computer program is executed by a processor, it can implement some or all of the steps in the above method, which will not be elaborated here.

[0163] In this article, "multiple" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0164] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a computer storage medium, which can be a computer-readable storage medium. When executed, the program can include the processes of the embodiments of the methods described above. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0165] The above-disclosed embodiments are merely some of the embodiments of this application, and should not be construed as limiting the scope of this application. Those skilled in the art can understand that all or part of the processes for implementing the above embodiments, and equivalent changes made in accordance with the claims of this application, still fall within the scope of this application.

Claims

1. A content recommendation method characterized by, The method includes: Obtain at least one content publishing account that has an interactive relationship with the target account during the historical period, and extract the first published content that meets the quality threshold from the published content of each content publishing account as the first candidate content set; Obtain at least one similar publishing account that has similar publishing attributes to each of the content publishing accounts. Based on the account exposure of the similar publishing accounts, obtain a target similar publishing account from the at least one similar publishing account. Then, obtain a second publishing content whose content quality reaches the quality threshold from the publishing content published by the target similar publishing account, as a second candidate content set. When constructing a target content set using the first candidate content set and the second candidate content set, the published content in the target content set is recommended to the target account.

2. The method of claim 1, wherein, The acquisition of at least one content-publishing account that has an interactive relationship with the target account within a historical period includes: Obtain first candidate posting accounts that have interacted with the target account; the account interactions include at least one of the following: following behavior, chat behavior; The interaction data performed by the target account against the first candidate posting account in N social scenarios during the historical period is obtained, and the scene interaction intensity between the target account and the first candidate posting account in each social scenario is determined based on the interaction data in the N social scenarios; N is a positive integer; The interaction strength between the target account and the first candidate publishing account is determined based on the interaction strength in the N social scenarios, and the first candidate publishing account with an interaction strength greater than the interaction strength threshold is regarded as at least one content publishing account that has the interaction relationship with the target account.

3. The method of claim 1, wherein, The acquisition of at least one content-publishing account that has an interactive relationship with the target account within a historical period includes: Based on the account selection strategy, a second candidate posting account with a potential association with the target account is selected from the account set; The interaction intensity is predicted between the target account and the second candidate posting account by using the interaction intensity processing model based on the account feature data of the target account and the account feature data of the second candidate posting account. The second candidate publishing account with an interaction intensity greater than the interaction intensity threshold is selected as at least one content publishing account that has the aforementioned interaction relationship with the target account; The account selection strategy includes at least one of the following: accounts that published content that the target account interacted with during the historical period, accounts associated with accounts that the target account interacted with during the historical period, accounts whose pages the target account viewed during the historical period, and accounts that were in the same group chat as the target account and whose group chat activity was greater than a preset activity level during the historical period.

4. The method of claim 3, wherein, The method further includes: Obtain a first sample account for training the initial interaction intensity processing model, and obtain a second sample account that has had account interactions with the first sample account; The interaction data performed by the first sample account against the second sample account in N social scenarios during the historical period is obtained, and the scene interaction intensity between the first sample account and the second sample account in each social scenario is determined based on the interaction data in the N social scenarios; N is a positive integer; The interaction intensity between the first sample account and the second sample account is determined based on the scene interaction intensity in the N social scenarios. The initial interaction intensity processing model is trained based on the interaction intensity of the sample accounts to obtain the trained interaction intensity processing model.

5. The method according to claim 4, characterized in that, The step of training the initial interaction intensity processing model based on the interaction intensity of the sample accounts to obtain the trained interaction intensity processing model includes: The initial interaction intensity processing model predicts the interaction intensity between the first sample account and the second sample account based on the account feature data of the first sample account and the account feature data of the second sample account, thereby obtaining the predicted account interaction intensity between the first sample account and the second sample account. Based on the interaction intensity of the sample accounts and the interaction intensity of the predicted accounts, the initial interaction intensity processing model is trained to obtain the trained interaction intensity processing model.

6. The method according to claim 1, characterized in that, The at least one content publishing account includes the target publishing account; The step of obtaining at least one similar posting account that has similar posting attributes to each of the content posting accounts includes: Select candidate similar accounts from the account set that have associated account tags with the target posting account; The posting attribute data of the target posting account is determined based on the interaction data and the posted content of the target posting account, and the posting attribute data of the candidate similar accounts is determined based on the interaction data and the posted content of the candidate similar accounts. The attribute similarity processing model is used to predict the attribute similarity between the target posting account and the candidate similar accounts based on the posting attribute data of the target posting account and the posting attribute data of the candidate similar accounts, so as to obtain the posting attribute similarity between the target posting account and the candidate similar accounts. Candidate similar accounts with a similarity of posting attributes greater than a similarity threshold are considered as at least one similar posting account that has similar posting attributes to the target posting account.

7. The method according to claim 1, characterized in that, The method of obtaining the target similar posting account from the at least one similar posting account based on the account exposure based on similar posting accounts includes: The account exposure of at least one similar posting account during the historical period is determined based on one or more of the following: number of followers, number of interactions with the posted content, and number of account views. The account whose exposure volume is greater than the exposure volume threshold among the at least one similar posting account is identified as the target similar posting account.

8. The method according to claim 1, characterized in that, The step of recommending content published in the target content set to the target account includes: Determine the content interaction rate of the target account for each piece of content published in the target content set; The target content set is sorted according to the content interaction rate to obtain the sorted target content set; Recommended content for the target account is selected from the sorted target content set and recommended to the target account.

9. The method according to claim 1, characterized in that, When constructing a target content set using the first candidate content set and the second candidate content set, recommending published content from the target content set to the target account includes: When a third candidate content set is obtained by recalling the published content set according to the content recall strategy, the first candidate content set, the second candidate content set, and the third candidate content set are determined as the target content set; Determine the content interaction rate of the target account for each piece of content published in the target content set; The target content set is sorted according to the content interaction rate to obtain the sorted target content set; Recommended content for the target account is selected from the sorted target content set and recommended to the target account; the recommended content includes at least M target content, and any one of the target content comes from the first candidate content set or the second candidate content set, where M is a positive integer.

10. The method according to claim 1, characterized in that, The target account refers to at least one of the following: an account that registers the target application for the first time within a preset time period, or an account that downloads and logs into the target application on a terminal device within a preset time period, then uninstalls the target application and re-downloads and logs into the target application.

11. A content recommendation device, characterized in that, The device includes: The content acquisition module is used to acquire at least one content publishing account that has an interactive relationship with the target account during the historical period, and to acquire the first published content that meets the quality threshold from the published content of each content publishing account, as the first candidate content set; The content acquisition module is further configured to acquire at least one similar publishing account that has similar publishing attributes to each content publishing account, acquire a target similar publishing account from the at least one similar publishing account based on the account exposure of the similar publishing account, and acquire second publishing content whose content quality reaches the quality threshold from the publishing content published by the target similar publishing account, as a second candidate content set; The content recommendation module is used to recommend published content from the target content set to the target account when a target content set is constructed using the first candidate content set and the second candidate content set.

12. An electronic device, characterized in that, The system includes a processor and a memory, wherein the memory is used to store a computer program, the computer program including program instructions, and the processor is configured to invoke the program instructions to perform the method as described in any one of claims 1-10.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the method as described in any one of claims 1-10.

14. A computer program product, characterized in that, The method includes a computer program comprising program instructions that, when executed by a processor, implement the method according to any one of claims 1-10.

Images