Content recommendation method, apparatus, device, storage medium, and program product

Abstract

This application discloses a content recommendation method, apparatus, device, storage medium, and program product, relating to the field of artificial intelligence. The method includes: acquiring a content recognition model, which is constructed based on first recommended content, the first recommended content being labeled high-quality recommended content, and the labeling process of the first recommended content not relying on the account's privacy information; recalling second recommended content from a recommended content library using the content recognition model, the second recommended content being high-quality recommended content identified by the content recognition model, and the recall process of the second recommended content not relying on the account's privacy information; and recommending content to the target account based on the first and second recommended content, provided the target account has disabled privacy information access permissions. The method provided in this application helps to increase the proportion of high-quality recommended content in the recalled content and improve the accuracy of content recommendation.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and in particular to a content recommendation method, apparatus, device, storage medium, and program product. Background Technology

[0002] Currently, in the content recommendation process, the first step is a recall process, which involves filtering candidate content from a massive amount of content. For example, in the process of recommending advertisements to users, the first step is to recall the advertisements.

[0003] In related technologies, during the content retrieval stage, users' relevant privacy information is often matched with content to obtain candidate content that the user may be interested in, which is then further filtered. However, this method relies on users' privacy information to achieve accurate retrieval. When users' privacy information cannot be obtained, the content retrieval volume and accuracy are low, affecting the content recommendation effect. Summary of the Invention

[0004] This application provides a content recommendation method, apparatus, device, storage medium, and program product, which helps to increase the proportion of high-quality recommended content in the recalled content and improve the accuracy of content recommendation. The technical solution is as follows:

[0005] On the one hand, embodiments of this application provide a content recommendation method, the method comprising:

[0006] A content recognition model is obtained, which is constructed based on a first recommended content, which is a high-quality recommended content that has been labeled, and the labeling process of the first recommended content does not depend on the privacy information of the account.

[0007] The content recognition model is used to recall second recommended content from the recommended content library. The second recommended content is high-quality recommended content identified by the content recognition model, and the recall process of the second recommended content does not depend on the privacy information of the account.

[0008] If the target account has disabled access to privacy information, content recommendations are made to the target account based on the first recommended content and the second recommended content.

[0009] On the other hand, embodiments of this application provide a content recommendation device, the device comprising:

[0010] The model acquisition module is used to acquire a content recognition model, which is constructed based on a first recommended content. The first recommended content is high-quality recommended content that has been labeled, and the labeling process of the first recommended content does not depend on the privacy information of the account.

[0011] The content recall module is used to recall second recommended content from the recommended content library through the content recognition model. The second recommended content is the identified high-quality recommended content, and the recall process of the second recommended content does not depend on the privacy information of the account.

[0012] The content recommendation module is used to recommend content to the target account based on the first recommended content and the second recommended content, when the target account has turned off the permission to access privacy information.

[0013] On the other hand, embodiments of this application provide a computer device including a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set, or instruction set, and the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the content recommendation method as described above.

[0014] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction, at least one program, code set, or instruction set is stored therein, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the content recommendation method as described above.

[0015] On the other hand, embodiments of this application provide a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the content recommendation method provided above.

[0016] The beneficial effects of the technical solutions provided in this application include at least the following:

[0017] In this embodiment, a content recognition model is constructed using the labeled first recommended content. This model can identify high-quality recommended content, thereby recalling high-quality second recommended content from the recommended content library. Since the process of labeling high-quality recommended content does not rely on the account's privacy information, and the process of recalling second recommended content is also done by the content recognition model, which does not rely on the account's privacy information, high-quality recommended content can be recalled without obtaining the account's privacy information during the content recall stage. This helps to increase the proportion of high-quality recommended content in the recalled content, thus enabling the recommendation of high-quality content to be sent to the target account even when the target account has closed its privacy information access permissions, thereby improving the accuracy of content recommendation. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A schematic diagram of an implementation environment provided by an exemplary embodiment of this application is shown;

[0020] Figure 2 A flowchart illustrating a content recommendation method provided in an exemplary embodiment of this application is shown;

[0021] Figure 3 A flowchart illustrating a content recommendation method provided by another exemplary embodiment of this application is shown;

[0022] Figure 4 A schematic diagram of a target ratio determination process provided in an exemplary embodiment of this application is shown;

[0023] Figure 5 This illustration shows an architectural diagram of a content recognition model for recognizing second recommended content, provided in an exemplary embodiment of this application.

[0024] Figure 6 A flowchart illustrating a content recommendation method provided by another exemplary embodiment of this application is shown;

[0025] Figure 7 A flowchart illustrating a content recommendation method provided by another exemplary embodiment of this application is shown;

[0026] Figure 8 This illustration shows a framework diagram of a content recommendation process provided by an exemplary embodiment of this application when the target account has enabled privacy access permissions;

[0027] Figure 9 This is a structural block diagram of a content recommendation apparatus provided in an exemplary embodiment of this application;

[0028] Figure 10 A schematic diagram of the structure of a computer device provided in an exemplary embodiment of this application is shown. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0030] 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.

[0031] In related technologies, content recommendation often involves matching user information with candidate content. However, there are situations where user privacy information cannot be obtained, making it impossible to recommend content based on user information and affecting the recommendation effect. Therefore, in this embodiment, high-quality recommended content obtained without relying on account privacy information will be used to retrieve content from the recommendation content library to obtain second recommended content. The retrieval process also does not rely on account privacy information, thus enabling the recommendation of high-quality content even when account privacy information cannot be obtained.

[0032] The method provided in this application can be applied to any content recommendation scenario. The following is an illustrative description of the application scenarios for the content recommendation method provided in this application.

[0033] 1. Applied to advertising recommendation scenarios

[0034] When applied to advertising recommendation scenarios, the method provided in this application embodiment can be applied to the backend server of an advertising push platform. The backend server can construct a content recognition model for identifying high-quality advertisements based on the labeled high-quality advertisements, recall advertisements in the advertisement library, obtain high-quality advertisements from the unlabeled advertisements, and then push advertisements to user accounts based on the high-quality advertisements.

[0035] 2. Applicable to video recommendation scenarios

[0036] When applied to video recommendation scenarios, the method provided in this application embodiment can be applied to the backend server of a video playback platform. The backend server can construct a content recognition model for identifying high-quality videos based on labeled high-quality videos, perform recall in the video library to obtain high-quality videos from unlabeled videos, and then recommend videos to user accounts based on these high-quality videos.

[0037] The above is only an illustrative example of the application scenario. The method provided in this application embodiment can also be applied to other scenarios that require content recommendation. This application embodiment does not limit the actual application scenario.

[0038] Figure 1A schematic diagram of an implementation environment provided by an exemplary embodiment of this application is shown. This implementation environment is illustrated using an advertising recommendation scenario as an example. The implementation environment includes a terminal 110 and a server 120. The terminal 110 and the server 120 communicate via a communication network. Optionally, the communication network can be a wired network or a wireless network, and the communication network can be at least one of a local area network (LAN), a metropolitan area network (MAN), and a wide area network (WAN).

[0039] Terminal 110 is an electronic device that provides advertising display functionality. This electronic device can be a mobile terminal such as a smartphone, tablet, or laptop computer, or a desktop computer, projector computer, etc. This application embodiment does not limit the specific type of terminal.

[0040] Server 120 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 databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. In this embodiment, server 120 is the server corresponding to the advertising push platform, which can determine the advertisements to be recommended to the target account and send the advertisements to terminal 110 for display.

[0041] like Figure 1 As shown, when a target account triggers a recommendation request, terminal 110 can send the request to server 120. Server 120 can determine the target account's privacy information access permission status. If the target account's privacy information access permission is off, it can determine the target advertisement to recommend to the target account based on the high-quality advertisements recalled in the advertisement library, and send the target advertisement to terminal 110 for display. Specifically, server 120 uses labeled high-quality advertisements to recall high-quality advertisements from the unlabeled advertisements in the advertisement library.

[0042] In another possible implementation, the entire content recommendation process described above can also be executed by terminal 110. That is, terminal 110 recalls high-quality advertisements from the advertisement library and recommends advertisements to target accounts. This application embodiment does not limit this. For ease of description, the following embodiments are described using the content recommendation method executed by a computer device as an example.

[0043] Please refer to Figure 2 The diagram illustrates a flowchart of a content recommendation method provided in an exemplary embodiment of this application. The method includes the following steps.

[0044] Step 201: Obtain the content recognition model. The content recognition model is constructed based on the first recommended content, which is high-quality recommended content that has been annotated, and the annotation process of the first recommended content does not rely on the privacy information of the account.

[0045] In related technologies, when recommending content to a target account, the process often relies on the target account's privacy information to determine recommended content that matches the target account. However, when the target account's privacy information cannot be obtained, the process of determining recommended content cannot rely on the account's privacy information, thus hindering accurate content recommendations. To achieve the goal of recommending high-quality content to a target account even when the account's privacy information cannot be obtained, in one possible implementation, the computer device first determines the first recommended content.

[0046] The first recommended content consists of high-quality recommendations flagged by computer devices based on content characteristics or recommendation results. This flagging process does not require the use of account privacy information. Optionally, the flagging process can be manual or automated based on relevant information from the recommended content.

[0047] Optional content recommendations can include advertisements, videos, articles, etc.

[0048] In one possible implementation, the computer device can use the first set of recommended content to construct a content recognition model for identifying high-quality recommended content. Then, the content recognition model can be used to identify high-quality recommended content within the unlabeled recommended content.

[0049] Step 202: Retrieve second recommended content from the recommended content library using a content recognition model. The second recommended content is the identified high-quality recommended content, and the retrieval process of the second recommended content does not rely on the account's privacy information.

[0050] In the recommendation process, content is first recalled from a massive amount of content. In this embodiment, during the recall phase, the computer device can obtain a content recognition model and use the content recognition model to recall second recommended content from the recommended content library.

[0051] In one possible implementation, the computer device can utilize a content recognition model to identify content in an unlabeled recommendation library, obtaining second recommendation content with similar characteristics to the first recommendation content, thereby obtaining high-quality recommendation content. During the recall process of the second recommendation content, it only utilizes the content recognition model based on the feature information of the second recommendation content for content recall, without relying on the account's privacy information.

[0052] Step 203: With the target account having disabled the permission to access privacy information, content recommendations are made to the target account based on the first and second recommended content.

[0053] In one possible implementation, a user may send a content recommendation request using a target account, for example, by searching for target content. When a computer device receives the content recommendation request, it will recommend relevant content to the target account based on the request.

[0054] During the content recommendation process, the computer device first determines its permission status to access the target account's privacy information. If it does not have permission to access the target account's privacy information, it recommends content to the target account based on high-quality recommended content. This high-quality recommended content includes the already labeled first recommended content and the recalled second recommended content. In one possible implementation, the computer device can further filter the first and second recommended content to obtain the final content recommended to the target account.

[0055] In one possible implementation, applications with recommendation functionality include controls to enable or disable privacy permissions. When a computer device determines that the privacy permission for a target account is disabled, it also determines that the target account's permission to access privacy information is disabled. In this case, the computer device uses high-quality recommended content, independent of account privacy information, to recommend content, thereby improving the accuracy of recommendations to the target account.

[0056] In summary, in this embodiment, a content recognition model is constructed using the labeled first recommended content. This model can identify high-quality recommended content, thereby recalling high-quality second recommended content from the recommended content library. Since the process of labeling high-quality recommended content does not rely on the account's privacy information, and the process of recalling second recommended content is also done by the content recognition model, which does not rely on the account's privacy information, high-quality recommended content can be recalled without obtaining the account's privacy information during the content recall stage. This helps to increase the proportion of high-quality recommended content in the recalled content, thus enabling the recommendation of high-quality content to be sent to the target account even when the target account has closed its privacy information access permissions, thereby improving the accuracy of content recommendation.

[0057] In this embodiment, a content recognition model is trained using the first recommended content, and then the trained content recognition model is used to identify high-quality recommended content in the recommended content library to obtain the second recommended content. The training process of the content recognition model will be described below with an exemplary embodiment.

[0058] Please refer to Figure 3 The diagram illustrates a flowchart of a content recommendation method provided by another exemplary embodiment of this application. The method includes the following steps.

[0059] Step 301: Mark the first recommended content in the recommended content library.

[0060] In this process, the computer device first labels the identified high-quality recommended content in the recommended content library to obtain the first recommended content. In one possible implementation, the labeling process can be based on feedback from the recommendation results. This feedback indicates the recommendation effect of the content after it has been recommended historically. Optionally, the feedback may include the click-through rate, conversion rate, page views, and revenue generated by the recommended content. The computer device can determine whether the content is high-quality based on the feedback and label it accordingly to obtain the first recommended content.

[0061] Optionally, the recommended content can be advertisements. When the recommended content is an advertisement, the first recommended content in the recommended advertisement library can be marked based on the advertisement recommendation requirements. Different advertisements have different advantages; in the process of determining high-quality recommended content, corresponding high-quality recommended advertisements can be identified according to the advertisement recommendation requirements, thus obtaining the first recommended content. Advertisement recommendation requirements refer to the advertisement recommendation effect required by the advertiser or advertising platform.

[0062] Optionally, if the advertising recommendation requirement is advertising conversion rate, ads in the recommended ad library with a total transaction GMV higher than the revenue threshold will be marked as the first recommended content.

[0063] Ad conversion rate is the percentage of viewers who make a purchase, register, or request information due to an ad, out of the total number of ad clicks. When the ad recommendation demand is ad conversion rate, computer devices can mark ads with a Gross Merchandise Volume (GMV) exceeding a revenue threshold from historically recommended ads as high-quality recommended content, i.e., designated as the primary recommended content. GMV is the product of ad bid and ad conversion volume; a higher GMV generally corresponds to a higher conversion rate. By designating ads with GMV exceeding the revenue threshold as primary recommended content, and then recalling secondary recommended content from the ad library based on such ads, the quality of ad recommendations can be ensured, thereby improving the conversion rate of recommended ads.

[0064] Optionally, if the advertising recommendation requirement is ad impressions, ads in the recommended ad library with more views than the viewing threshold will be marked as the first recommended content.

[0065] In another possible scenario, the advertising recommendation demand is for ad impressions. In this case, the computer device can identify ads with more views than the viewing threshold among the recommended ads as the first recommended content, so that the recalled second recommended content will have a higher exposure after being recommended.

[0066] Optionally, if the advertising recommendation demand breaks even with the advertising cost, ads in the recommendation ad library with GMV higher than the advertising cost investment will be designated as the first recommended content.

[0067] In another possible scenario, the demand for ad recommendations might be to break even on advertising costs, where breaking even means that advertising revenue is greater than or equal to advertising costs. Therefore, computer equipment can identify ads with a GMV greater than their advertising costs as the primary recommended content, thus ensuring that the cost-effectiveness of the advertising is achieved after the recall of secondary recommended content.

[0068] The above describes the process of labeling based on the feedback of recommendation results. In another possible implementation, the computer device can also label the recommended content based on its content attributes. For example, when the recommended content is an advertisement, the content attributes may include advertising investment, advertising creator, and products featured in the advertisement; when the recommended content is a video or article, the content attributes may include the content author and the meaning of the content; when the recommended content is a product, the content attributes may include the product price, quality, and brand. This application does not limit the labeling method.

[0069] Step 302: Based on the first recommended content and the unlabeled candidate recommended content in the recommended content library, determine the first negative sample content.

[0070] In addition to the definitively identified high-quality recommended content, the recommended content library also contains undetermined content whose quality cannot be determined based on current information. Therefore, in this embodiment, a content recognition model will be trained using labeled high-quality recommended content to identify content among unlabeled candidate recommended content. Optionally, the content recognition model can be a Deep Neural Networks (DNN) classifier.

[0071] When training a content recognition model, both positive and negative samples are required. Computer devices can only obtain labeled, high-quality recommended content as positive samples, but cannot obtain accurate negative samples for model training. Therefore, in one possible implementation, the computer device first needs to determine reliable negative sample content, which, along with the positive sample content, will be used to train the content recognition model. The process of determining the first negative sample content may include the following steps 302a-302c (not shown in the figure):

[0072] Step 302a: Select target sample content from the first recommended content.

[0073] In one possible implementation, the computer device selects a portion of the positive samples as negative samples to identify the content of the first negative sample. First, target sample content is selected from the first recommended content; optionally, the target sample content can be recommended content at a target proportion within the first recommended content.

[0074] The target ratio can be a pre-set ratio, and the size of the target ratio will affect the content of the final determined first negative sample. Therefore, in one possible implementation, the computer device will pre-determine the size of the target ratio using a grid search experiment.

[0075] Optionally, the computer device can adjust the scale according to the target step size and verify the recommendation effectiveness of the first negative sample determined based on the adjusted scale after each adjustment. After multiple adjustments, the scale corresponding to the lowest recommendation effectiveness of the first negative sample is selected as the target scale. For example, taking advertising content as the recommended content, the computer device gradually adjusts the scale in 5% steps and determines the recommendation effectiveness of the first negative sample content under various scales. The recommendation effectiveness is determined based on the ad's click-through rate (CTCVR). Figure 4 As shown, it illustrates the recommendation effect corresponding to the first negative sample content determined at different proportions. It can be seen that when the proportion is 20%, the average CTCVR of the first negative sample content / the average CTCVR of the candidate recommended content is the lowest. Therefore, the first negative sample content identified is most accurate when the proportion is 20%. Thus, the target proportion can be set at 20%.

[0076] Step 302b: Use the second positive sample content and the second negative sample content to train the content recognition model. The second positive sample content is the recommended content in the first recommended content excluding the target sample content. The second negative sample content includes candidate recommended content and target sample content.

[0077] After the computer device selects the target sample content of the target proportion, it can determine the second positive sample content and the second negative sample content based on the target sample content, thereby training the content recognition model.

[0078] The computer device uses the recommended content in the first recommended content P, excluding the target sample content S, as the second positive sample content, i.e., the second positive sample content is PS; and uses the unlabeled candidate recommended content U and the target sample content S in the recommended content library as the second negative sample content, i.e., the second negative sample content is U+S, to train the content recognition model.

[0079] Optionally, the model parameters of the content recognition model can be updated using backpropagation or gradient update. The content recognition model is then used to identify the second positive and second negative sample content to obtain content prediction probabilities. Based on the predicted sample categories indicated by these probabilities and the corresponding preset sample categories, the content recognition model is trained using backpropagation to obtain the trained content recognition model. The predicted sample category is determined based on whether it belongs to the second positive or second negative sample content. When it belongs to the second positive sample content, its preset sample category is high-quality recommended content; when it belongs to the second negative sample content, its preset sample category is low-quality recommended content.

[0080] Step 302c: Using a content recognition model, identify the content of the first negative sample in the content of the second negative sample.

[0081] After training the content recognition model using the second positive sample content and the second negative sample content, the content recognition model can be used to identify each second negative sample content, and the first negative sample content can be determined based on the identification results. This process may include the following steps:

[0082] Step 1: Input the content of the second negative sample into the content recognition model to obtain the content prediction probability of each second negative sample. The content prediction probability is used to indicate the probability that the recommended content is high-quality recommended content.

[0083] In one possible implementation, the computer device uses a trained content recognition model to identify the content of each second negative sample and obtain the corresponding content prediction probability.

[0084] Step 2: If the content prediction probability is lower than the probability threshold, the second negative sample content corresponding to the content prediction probability is determined as the first negative sample content. The probability threshold is the minimum value of the content prediction probability corresponding to the target sample content.

[0085] After the computer device obtains the content prediction probabilities of each second negative sample content, the content prediction probabilities of the target sample content in the second negative sample content are sorted, and the minimum content prediction probability of the target sample content is obtained and used as the probability threshold. For example, if the minimum content prediction probability of the target sample content is 30%, then 30% can be determined as the probability threshold.

[0086] Then, the computer equipment determines the predicted probability of each unlabeled candidate recommendation content, and identifies the candidate recommendation content with a probability lower than the probability threshold as the first negative sample content, thus obtaining the reliable negative sample obtained from the identification.

[0087] Step 303: Use the first positive sample content and the first negative sample content to train the content recognition model. The first positive sample content includes the first recommended content.

[0088] The process described above, which uses the content of the second positive sample and the content of the second negative sample to train the content recognition model, is used to identify reliable negative samples. After identifying reliable negative samples (the content of the first negative sample), the computer device uses the content of the first positive sample and the content of the first negative sample to retrain the content recognition model, thereby identifying high-quality recommended content.

[0089] Optionally, the first positive sample content includes all recommended content from the first recommended content.

[0090] Similarly, the model parameters of the content recognition model can be updated using backpropagation or gradient update. The content recognition model is then used to identify the first positive and first negative sample content to obtain content prediction probabilities. Based on the predicted sample categories indicated by these probabilities and the true sample categories, the content recognition model is trained using backpropagation to obtain the trained content recognition model. The true sample category is determined by whether it belongs to the first positive or first negative sample content. When it belongs to the first positive sample content, its true sample category is high-quality recommended content; when it belongs to the first negative sample content, its true sample category is low-quality recommended content.

[0091] Step 304: Obtain the content recognition model.

[0092] After the content recognition model is trained on the computer device, the model can be stored. When content retrieval is needed, the pre-trained content recognition model can be retrieved.

[0093] Step 305: Retrieve the second recommended content from the recommended content library using the content recognition model.

[0094] After training the content recognition model using the first positive sample content and the first negative sample content, the computer device can use the content recognition model to identify candidate recommendation content in the recommendation content library and obtain the high-quality recommendation content of the candidate recommendation content, that is, the second recommendation content.

[0095] In one possible implementation, the computer device can use a content recognition model to determine the predicted content probability of each candidate recommendation, and identify the candidate recommendation with a predicted content probability greater than a target probability as the second recommendation. For example, the target probability is 80%.

[0096] This is an illustrative example of the structure used to recall second-recommended content using a content recognition model. Figure 5As shown, firstly, reliable negative samples 503 are identified using the first recommended content 501 and the unlabeled candidate recommended content 502. Then, the content recognition model 504 is trained using the first recommended content 501 as a positive sample and the identified reliable negative samples 503. Based on the trained content recognition model 504, the second recommended content is recalled from the recommended content library.

[0097] Step 306: With the target account having disabled the permission to access privacy information, content recommendations are made to the target account based on the first and second recommended content.

[0098] The implementation method of this step can refer to step 203 above, and will not be repeated here.

[0099] In this embodiment, the labeled high-quality recommendation content and the unlabeled candidate recommendation content are used to determine reliable negative samples for training. Then, the labeled high-quality recommendation content is used as positive samples to train the content recognition model, thereby improving the accuracy of the content recognition model in recognizing high-quality recommendation content.

[0100] In this embodiment, corresponding high-quality recommended ads are determined according to different advertising needs, so that the recalled high-quality recommended ads can meet the advertising needs and ensure the effectiveness of advertising recommendations.

[0101] After recalling the second recommended content, the computer device can recommend content to the target account based on the first and second recommended content. During this process, the first and second recommended content need to be further filtered, and the final content recommendation is based on the filtering results. The following will describe the process with an exemplary embodiment.

[0102] Please refer to Figure 6 The diagram illustrates a flowchart of a content recommendation method provided by another exemplary embodiment of this application. The method includes the following steps.

[0103] Step 601: Obtain the content recognition model.

[0104] The implementation method for this step can be referred to the above embodiments, and will not be repeated in this embodiment.

[0105] Step 602: Retrieve the second recommended content from the first candidate content in the recommended content library using the content recognition model.

[0106] The recommendation content library contains multiple candidate recommendation contents. In one possible implementation, when the computer device recalls content from the recommendation content library, it recalls different candidate recommendation contents every target period, thereby making content recommendations based on different recall results in different periods and updating the recommendation content in a timely manner.

[0107] The computer device first recalls the second recommended content from the first candidate content. The first candidate content is a portion of the recommended content library.

[0108] Step 603: With the target account having disabled the permission to access privacy information, the first recommended content and the second recommended content are filtered based on the first targeting condition to obtain the first targeted recommended content.

[0109] The process of targeted filtering based on the first targeting condition does not rely on the account's privacy information. The first targeting condition includes the recommendation restrictions for the first recommended content and the second recommended content.

[0110] If the target account has disabled access to privacy information, the computer device will recommend content based on a first set of recommended content and a second set of recommended content. However, since both sets of recommended content still contain a large amount of content, further filtering is required. This first involves a targeted filtering process, where filtering is based on a first targeting condition. Different recommended content has different recommendation restrictions, i.e., targeting conditions. Targeting conditions can be set by the content provider, and can include restrictions on recommendation time, region, and duration. For example, when the recommended content is an advertisement, the targeting conditions can be pre-set by the advertiser, such as region, time, and target audience.

[0111] Computer devices can perform targeted filtering based on a first targeting condition. This targeted filtering process does not rely on account privacy information. The first targeting condition includes the targeting conditions for each recommended content within the first and second recommended content sets. During targeted filtering, it can be determined whether a recommendation can be sent to the target account based on the corresponding recommendation restrictions for each recommended content. In other words, based on the first targeting condition, recommended content whose current recommendation status does not match the targeting condition is filtered out. For example, if the targeting condition for recommended content A is a time-based condition, limiting its delivery to 19:00-21:00, but the recommended time to the target account is 13:00, then recommended content A cannot be recommended to the target account and must be filtered.

[0112] Furthermore, during the filtering process based on the first targeting condition, targeted filtering is only performed based on targeting conditions that do not require access to account privacy information. For example, the targeting conditions for recommended content A include two types: recommendation time and recommendation target. The recommendation target refers to recommending content A to accounts that are interested in the corresponding type of content. When filtering based on the recommendation target, it is necessary to obtain the account's privacy information. However, the recommendation time does not require access to the account's privacy information. Therefore, targeted filtering is performed only based on the recommendation time condition.

[0113] Furthermore, targeted filtering can include a non-breakable targeted filtering process. Non-breakable targeted filtering refers to selecting recommended content that meets necessary targeting conditions, which are conditions that must be met. For example, when the recommended content is an advertisement, the necessary targeting conditions are the conditions that the advertiser must set for the ad placement.

[0114] In one possible implementation, the computer device may perform unbreakable targeted filtering on only the first recommended content and the second recommended content to obtain the first targeted recommended content.

[0115] Step 604: Based on the priority of the first targeted recommended content, sort the first targeted recommended content to obtain the first recommended content sequence, wherein the priority of the second recommended content in the first targeted recommended content is higher than the priority of the first recommended content.

[0116] In one possible implementation, after obtaining the first targeted recommendation content, the first targeted recommendation content can be sorted. The sorting process includes aggregate optimization, coarse ranking, and fine ranking. The aggregate optimization process can involve deduplicating similar recommendation content, followed by coarse ranking and fine ranking to obtain the quality scores corresponding to each of the first targeted recommendation content, thereby obtaining the first recommendation content sequence based on the quality scores.

[0117] Optionally, since the second recommended content is the most recently identified high-quality recommended content, recommending based on the second recommended content ensures the timeliness of the recommended content. During the ranking of the first targeted recommended content, the second recommended content has a higher ranking priority than the first recommended content; that is, the second recommended content is recommended first. Furthermore, the corresponding quality can be determined based on the content prediction probability of the second recommended content identified by the content recognition model, thereby ranking the second recommended content based on its quality. The content prediction probability is positively correlated with the quality; that is, the higher the content prediction probability, the higher the quality of the corresponding recommended content, and the higher its ranking priority.

[0118] Correspondingly, the process of ranking the first targeted recommended content also does not rely on the account's privacy information.

[0119] Step 605: Based on the first recommended content sequence, recommend content to the target account.

[0120] The computer device determines the target number of recommended content items that are ranked first in the first recommended content sequence, and then recommends content to the target account.

[0121] Step 606: When the target period is reached, the fourth recommended content is recalled from the second candidate content in the recommended content library through the content recognition model. The fourth recommended content is used to replace the first and second recommended content. The second candidate content is different from the first candidate content.

[0122] In one possible implementation, after the target period is reached, the computer device will rotate the content used for content recommendation. Specifically, using a content recognition model, content is recalled from another portion of the candidate recommendation content (second candidate content) in the recommendation content library to obtain fourth recommended content. When subsequently recommending content to the target account, the computer device can use the recalled fourth recommended content, along with the first and second recommended content, to update the recommended content. Alternatively, after recalling the fourth recommended content, subsequent content recommendations may be based solely on the fourth recommended content, replacing the first and second recommended content to ensure the timeliness of the recommended content. This embodiment does not limit this approach.

[0123] Furthermore, in one possible implementation, high-quality recommended content can be re-labeled at fixed intervals. The content recognition model can then be updated using the re-labeled high-quality recommended content, and the high-quality recommended content can be identified from the recommended content library based on the updated content recognition model, thus ensuring the accuracy of the high-quality recommended content.

[0124] In this embodiment, content is recalled from different candidate recommendation content sets every target period to update the recommendation content in stages and ensure the timeliness of the recommendation content.

[0125] The above embodiments illustrate how to recommend content when the target account has disabled privacy information access permissions. In another possible scenario, the target account may have enabled privacy information access permissions. The following embodiments will exemplify how to recommend content when the target account has enabled privacy information access permissions.

[0126] Please refer to Figure 7 The diagram illustrates a flowchart of a content recommendation method provided by another exemplary embodiment of this application. The method includes the following steps.

[0127] Step 701: Obtain the content recognition model.

[0128] Step 702: Retrieve the second recommended content from the recommended content library using the content recognition model.

[0129] Step 703: With the target account having disabled the permission to access privacy information, content recommendations are made to the target account based on the first and second recommended content.

[0130] The implementation methods for steps 701 to 703 can be referred to the above embodiments, and will not be repeated in this embodiment.

[0131] Step 704: If the target account has enabled permission to access privacy information, determine the account profile based on the target account's privacy information.

[0132] When a computer device determines that the target account's privacy information is enabled, the computer device can construct an account profile of the target account based on relevant information about the target account. This account profile represents the account's interest characteristics. In one possible implementation, the computer device can determine the account profile based on the target account's historical behavior information, etc. Optionally, the account profile can be represented using tags.

[0133] Step 705: Retrieve third recommended content from the recommended content library based on the account profile. The third recommended content matches the account profile.

[0134] In this system, the computer device can retrieve content from the recommendation content library based on the target account's account profile to obtain third-party recommended content. In one possible implementation, the computer device uses a K-Nearest Neighbor (KNN) model to match the account profile with candidate recommended content. Specifically, it matches the tags corresponding to the candidate recommended content with the tags indicated by the account profile, thereby obtaining the third-party recommended content. This third-party recommended content is the personalized recommendation content matched to the account profile.

[0135] Step 706: Based on the first recommended content, the second recommended content, and the third recommended content, content recommendations are made to the target account.

[0136] After receiving the third recommended content, the computer device can make content recommendations based on the first, second, and third recommended content. The first and second recommended content are non-personalized recommendations unrelated to the target account, while the third recommended content is personalized recommendations matched to the target account. That is, when the target account has granted permission to access privacy information, the computer device makes content recommendations based on both personalized and non-personalized recommendations. This method may include the following steps 706a-706c:

[0137] Step 706a: Based on the second targeting conditions, perform targeted filtering on the first recommended content, the second recommended content, and the third recommended content to obtain the second targeted recommended content. The second targeting conditions include the recommendation restrictions of the first recommended content, the second recommended content, and the third recommended content.

[0138] Accordingly, in the content recommendation process based on the first, second, and third recommended content, targeted filtering is first required. In this case, the computer device merges the first, second, and third recommended content into a content queue and then performs targeted filtering on the merged content sequence. The targeted filtering based on the second targeting condition is done in the same way as the targeted filtering based on the first targeting condition, used to filter out recommended content whose current recommendation status does not match the targeting conditions. However, during the targeted filtering based on the second targeting condition, because the account's privacy information can be obtained, filtering can be performed based on all targeting conditions corresponding to the recommended content to obtain the second targeted recommended content.

[0139] Step 706b: Sort the second targeted recommendation content to obtain the second recommendation content sequence, wherein the third recommendation content has a higher priority than the first recommendation content and the second recommendation content.

[0140] After targeted filtering, the second targeted recommendation content is sorted to obtain a second recommendation content sequence. Since the third recommendation content is personalized content retrieved based on account profiles, while the first and second recommendation content are non-personalized, personalized recommendation content has a higher degree of matching with the target account than non-personalized recommendation content, meaning it is more likely to be of interest to the user. Therefore, in one possible implementation, during the sorting process of the second targeted recommendation content, the third recommendation content has a higher priority than the first and second recommendation content; that is, the third recommendation content can be placed before the first and second recommendation content. The sorting method for the first and second recommendation content can be the same as in step 604 above, and will not be repeated in this embodiment.

[0141] Step 706c: Based on the second recommended content sequence, recommend content to the target account.

[0142] Optionally, the computer device recommends content to the target account based on a second recommended content sequence.

[0143] To illustrate, taking recommended content as an example, assuming the target account has enabled privacy information access permissions, the overall recall framework is as follows: Figure 8As shown, when recalling ads based on account profiles, recall can be performed in both the full-database ads and the expanded-scale ads. Specifically, in the full-database ads 801, the third recommended content is retrieved based on the account profile; in the non-personalized ad database 802, the second recommended content is retrieved based on the first recommended content (i.e., the content recognition model built based on the first recommended content identifies the second recommended content in the non-personalized ad database 802); and in the expanded-scale ads 803, the third recommended content is retrieved based on the account profile. The expanded-scale ads 803 can specify ads for some advertisers in the full-database ads 801. Afterwards, a targeted filtering process is performed (including non-breakable targeted filtering of ads retrieved from the non-personalized ad database and ads retrieved from the expanded-scale ads), and the filtered ads are merged to obtain a second targeted ad sequence. Then, the second targeted ad sequence undergoes pre-sorting, coarse-sorting, and fine-sorting processes to obtain the final recommended ad sequence.

[0144] In this embodiment, when the target account has granted permission to access account privacy information, the computer device can construct an account profile based on the privacy information and retrieve content based on the account profile to obtain third recommended content that matches the account profile. Thus, content recommendation is performed based on the first recommended content, the second recommended content, and the third recommended content, further improving the accuracy of content recommendation.

[0145] It should be noted that all information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.), and signals involved in this application have been authorized by the user or fully authorized by all parties, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, the account privacy information involved in this application was obtained with full authorization.

[0146] Figure 9 This is a structural block diagram of a content recommendation apparatus provided in an exemplary embodiment of this application. As shown in the figure, the apparatus includes:

[0147] The model acquisition module 901 is used to acquire a content recognition model, which is constructed based on a first recommended content. The first recommended content is high-quality recommended content that has been labeled, and the labeling process of the first recommended content does not depend on the privacy information of the account.

[0148] The content retrieval module 902 is used to retrieve second recommended content from the recommended content library through the content recognition model. The second recommended content is the identified high-quality recommended content, and the retrieval process of the second recommended content does not depend on the privacy information of the account.

[0149] The content recommendation module 903 is used to recommend content to the target account based on the first recommended content and the second recommended content when the target account has closed its privacy information access permission.

[0150] Optionally, the device further includes:

[0151] The annotation module is used to annotate the first recommended content in the recommended content library;

[0152] The content determination module is used to determine the first negative sample content based on the first recommended content and the unlabeled candidate recommended content in the recommended content library;

[0153] The model training module is used to train the content recognition model using the first positive sample content and the first negative sample content, wherein the first positive sample content includes the first recommended content.

[0154] Optionally, the content determination module is further configured to:

[0155] Select target sample content from the first recommended content;

[0156] The content recognition model is trained using the second positive sample content and the second negative sample content. The second positive sample content is the recommended content in the first recommended content excluding the target sample content, and the second negative sample content includes the candidate recommended content and the target sample content.

[0157] Using the content recognition model, the first negative sample content is identified in the second negative sample content.

[0158] Optionally, the content determination module is further configured to:

[0159] The second negative sample content is input into the content recognition model to obtain the content prediction probability of each second negative sample content. The content prediction probability is used to indicate the probability that the recommended content is the high-quality recommended content.

[0160] If the predicted content probability is lower than the probability threshold, the second negative sample content corresponding to the predicted content probability is determined as the first negative sample content, where the probability threshold is the minimum value of the predicted content probability corresponding to the target sample content.

[0161] Optionally, the content recommendation module 903 is further used for:

[0162] Based on the first targeting condition, the first recommended content and the second recommended content are targeted and filtered to obtain the first targeted recommended content. The targeted filtering process based on the first targeting condition does not rely on the privacy information of the account. The first targeting condition includes the recommendation restriction conditions of the first recommended content and the second recommended content.

[0163] Based on the arrangement priority of the first targeted recommendation content, the first targeted recommendation content is sorted to obtain a first recommendation content sequence, wherein the arrangement priority of the second recommendation content in the first targeted recommendation content is higher than the arrangement priority of the first recommendation content;

[0164] Based on the first recommended content sequence, content is recommended to the target account.

[0165] Optionally, the device further includes:

[0166] The profile determination module is used to determine an account profile based on the privacy information of the target account when the target account has enabled the permission to obtain privacy information.

[0167] The content recall module 902 is also used to recall third recommended content in the recommended content library based on the account profile, wherein the third recommended content matches the account profile;

[0168] The content recommendation module 903 is further configured to recommend content to the target account based on the first recommended content, the second recommended content, and the third recommended content.

[0169] Optionally, the content recommendation module 903 is further used for:

[0170] Based on the second targeting condition, the first recommended content, the second recommended content, and the third recommended content are targeted and filtered to obtain the second targeted recommended content. The second targeting condition includes the recommendation restriction conditions of the first recommended content, the second recommended content, and the third recommended content.

[0171] Based on the ranking priority of the second targeted recommendation content, the second targeted recommendation content is sorted to obtain a second recommendation content sequence, wherein the ranking priority of the third recommendation content in the second targeted recommendation content is higher than the ranking priority of the first recommendation content and the second recommendation content;

[0172] Based on the second recommended content sequence, content is recommended to the target account.

[0173] Optionally, the content retrieval module 902 is further configured to:

[0174] The content recognition model is used to recall the second recommended content from the first candidate content in the recommended content library.

[0175] When the target period is reached, the content recognition model recalls a fourth recommended content from the second candidate content in the recommended content library. The fourth recommended content is used to replace the first recommended content and the second recommended content. The second candidate content is different from the first candidate content.

[0176] Optional, recommended content is advertising;

[0177] The annotation module is also used for:

[0178] Based on the advertising recommendation requirements, the first recommended content in the recommended advertising library is labeled.

[0179] Optionally, the annotation module is further configured to:

[0180] When the advertising recommendation requirement is advertising conversion rate, advertisements in the recommended advertising library whose total transaction GMV is higher than the revenue threshold are marked as the first recommended content;

[0181] When the advertising recommendation requirement is advertising exposure, advertisements in the recommended advertising library with a view count higher than the view threshold are marked as the first recommended content;

[0182] If the advertising recommendation demand is equal to the advertising cost, then the advertisements in the recommended advertisement library whose GMV is higher than the advertising cost are marked as the first recommended content.

[0183] In summary, in this embodiment, a content recognition model is constructed using the labeled first recommended content. This model can identify high-quality recommended content, thereby recalling high-quality second recommended content from the recommended content library. Since the process of labeling high-quality recommended content does not rely on the account's privacy information, and the process of recalling second recommended content is also done by the content recognition model, which does not rely on the account's privacy information, high-quality recommended content can be recalled without obtaining the account's privacy information during the content recall stage. This helps to increase the proportion of high-quality recommended content in the recalled content, thus enabling the recommendation of high-quality content to be recommended to the target account even when the target account has closed its privacy information access permissions, thereby improving the accuracy of recommendations.

[0184] It should be noted that the apparatus provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the apparatus can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and their implementation process can be found in the method embodiments, which will not be repeated here.

[0185] Please refer to Figure 10 This illustration shows a schematic diagram of the structure of a computer device provided in an exemplary embodiment of this application. Specifically, the computer device 1000 includes a Central Processing Unit (CPU) 1001, a system memory 1004 including a random access memory 1002 and a read-only memory 1003, and a system bus 1005 connecting the system memory 1004 and the CPU 1001. The computer device 1000 also includes a basic input / output system (I / O system) 1006 that facilitates the transfer of information between various devices within the computer, and a mass storage device 1007 for storing the operating system 1013, application programs 1014, and other program modules 1015.

[0186] In some embodiments, the basic input / output system 1006 includes a display 1008 for displaying information and an input device 1009 for user input, such as a mouse or keyboard. Both the display 1008 and the input device 1009 are connected to the central processing unit 1001 via an input / output controller 1010 connected to the system bus 1005. The basic input / output system 1006 may also include the input / output controller 1010 for receiving and processing input from multiple other devices such as a keyboard, mouse, or electronic stylus. Similarly, the input / output controller 1010 also provides output to a display screen, printer, or other types of output devices.

[0187] The mass storage device 1007 is connected to the central processing unit 1001 via a mass storage controller (not shown) connected to the system bus 1005. The mass storage device 1007 and its associated computer-readable media provide non-volatile storage for the computer device 1000. That is, the mass storage device 1007 may include computer-readable media (not shown) such as a hard disk or drive.

[0188] Without loss of generality, the computer-readable medium may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented using any method or technology for storing information such as computer-readable instructions, data structures, program modules, or other data. Computer storage media include random access memory (RAM), read-only memory (ROM), flash memory or other solid-state storage technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic tape cassettes, magnetic tape, disk storage, or other magnetic storage devices. Of course, those skilled in the art will recognize that the computer storage media are not limited to the above-mentioned types. The system memory 1004 and mass storage device 1007 described above can be collectively referred to as memory.

[0189] The memory stores one or more programs, which are configured to be executed by one or more central processing units 1001. The one or more programs contain instructions for implementing the methods described above, and the central processing unit 1001 executes the one or more programs to implement the methods provided in the various method embodiments described above.

[0190] According to various embodiments of this application, the computer device 1000 can also be connected to a remote computer on a network, such as the Internet. That is, the computer device 1000 can be connected to a network 1012 via a network interface unit 1011 connected to the system bus 1005, or it can use a network interface unit 1010 to connect to other types of networks or remote computer systems (not shown).

[0191] The memory further includes one or more programs stored in the memory, and the one or more programs include steps performed by a computer device in the methods provided in the embodiments of this application.

[0192] This application also provides a computer-readable storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement the content recommendation method described in any of the above embodiments.

[0193] This application provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the content recommendation method provided above.

[0194] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. This program can be stored in a computer-readable storage medium, which may be a computer-readable storage medium included in the memory described in the above embodiments; or it may be a standalone computer-readable storage medium not assembled into a terminal. The computer-readable storage medium stores at least one instruction, at least one program segment, a code set, or an instruction set. The at least one instruction, the at least one program segment, the code set, or the instruction set is loaded and executed by a processor to implement the content recommendation method described in any of the above method embodiments.

[0195] Optionally, the computer-readable storage medium may include ROM, RAM, solid-state drives (SSDs), or optical discs, etc. The RAM may include resistive random access memory (ReRAM) and dynamic random access memory (DRAM). The sequence numbers of the embodiments described above are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0196] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0197] The above description is merely an optional embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A content recommendation method characterized by, The method comprises: annotating first recommended content in a recommended content library, the first recommended content being high-quality recommended content that has been annotated, and the annotation process of the first recommended content not depending on privacy information of an account; selecting target sample content in the first recommended content; training a content recognition model using second positive sample content and second negative sample content, the second positive sample content being recommended content in the first recommended content other than the target sample content, and the second negative sample content including candidate recommended content and the target sample content; after training the content recognition model using the second positive sample content and the second negative sample content is complete, identifying first negative sample content in the second negative sample content using the content recognition model; training the content recognition model using first positive sample content and the first negative sample content, the first positive sample content including the first recommended content; obtaining the content recognition model; recalling second recommended content from the recommended content library through the content recognition model, the second recommended content being high-quality recommended content identified by the content recognition model, and the recall process of the second recommended content not depending on privacy information of an account; in a content recommendation process, determining a permission situation of obtaining privacy information of a target account; in a case where the target account closes the permission of obtaining privacy information, performing content recommendation on the target account based on the first recommended content and the second recommended content.

2. The method of claim 1, wherein, The use of the content recognition model to identify first negative sample content in the second negative sample content comprises: inputting the second negative sample content into the content recognition model to obtain content prediction probability of each second negative sample content, the content prediction probability being used to indicate a probability that recommended content is the high-quality recommended content; in a case where the content prediction probability is lower than a probability threshold value, determining the second negative sample content corresponding to the content prediction probability as the first negative sample content, the probability threshold value being a minimum value of the content prediction probability corresponding to the target sample content.

3. The method according to any one of claims 1 to 2, characterized in that, The performance of content recommendation on the target account based on the first recommended content and the second recommended content comprises: performing directional filtering on the first recommended content and the second recommended content based on a first directional condition to obtain first directional recommended content, wherein the process of directional filtering based on the first directional condition does not depend on privacy information of an account, and the first directional condition includes a recommendation restriction condition of the first recommended content and the second recommended content; sorting the first directional recommended content based on arrangement priority of the first directional recommended content to obtain a first recommended content sequence, wherein the arrangement priority of the second recommended content in the first directional recommended content is higher than the arrangement priority of the first recommended content; performing content recommendation on the target account based on the first recommended content sequence.

4. The method according to any one of claims 1 to 2, characterized in that, The method further comprises: in a case where the target account opens the permission of obtaining privacy information, determining an account portrait based on privacy information of the target account; recalling third recommendation content in the recommendation content library based on the account portrait, the third recommendation content matching the account portrait; performing content recommendation on the target account based on the first recommendation content, the second recommendation content, and the third recommendation content.

5. The method of claim 4, wherein, The content recommendation on the target account based on the first recommendation content, the second recommendation content, and the third recommendation content comprises: performing directional filtering on the first recommendation content, the second recommendation content, and the third recommendation content based on a second directional condition to obtain second directional recommendation content, the second directional condition including a recommendation restriction condition of the first recommendation content, the second recommendation content, and the third recommendation content; performing sorting on the second directional recommendation content based on an arrangement priority of the second directional recommendation content to obtain a second recommendation content sequence, wherein the arrangement priority of the third recommendation content in the second directional recommendation content is higher than the arrangement priority of the first recommendation content and the second recommendation content; performing content recommendation on the target account based on the second recommendation content sequence.

6. The method of any one of claims 1 to 2, wherein, The recalling of the second recommendation content from the recommendation content library by the content recognition model comprises: recalling the second recommendation content from first candidate content in the recommendation content library by the content recognition model; The method further comprises: in the case of reaching a target period, recalling fourth recommendation content from second candidate content in the recommendation content library by the content recognition model, the fourth recommendation content being used to replace the first recommendation content and the second recommendation content, and the second candidate content being different from the first candidate content.

7. The method of any one of claims 1 to 2, wherein, The recommendation content is an advertisement. The labeling of the first recommendation content in the recommendation content library comprises: labeling the first recommendation content in a recommendation advertisement library based on an advertisement recommendation demand.

8. The method of claim 7, wherein, The labeling of the first recommendation content in the recommendation advertisement library based on the advertisement recommendation demand comprises: in the case of the advertisement recommendation demand being an advertisement conversion rate, labeling an advertisement with a gross merchandise volume (GMV) higher than a yield threshold in the recommendation advertisement library as the first recommendation content; in the case of the advertisement recommendation demand being an advertisement exposure amount, labeling an advertisement with a browsing volume higher than a browsing threshold in the recommendation advertisement library as the first recommendation content; in the case of the advertisement recommendation demand being advertisement cost leveling, labeling an advertisement with the GMV higher than an advertisement cost input in the recommendation advertisement library as the first recommendation content.

9. A content recommendation apparatus characterized by comprising: The device comprises: a labeling module configured to label first recommendation content in a recommendation content library, the first recommendation content being high-quality recommendation content after labeling, and a labeling process of the first recommendation content not depending on privacy information of an account; a content determination module configured to select target sample content in the first recommendation content; a content recognition model is trained using second positive sample content and second negative sample content, the second positive sample content being recommendation content in the first recommendation content except the target sample content, and the second negative sample content including candidate recommendation content and the target sample content. After the content recognition model is trained by using the second positive sample content and the second negative sample content, the content recognition model is used to identify first negative sample content in the second negative sample content; The model training module is configured to train the content recognition model by using first positive sample content and the first negative sample content, the first positive sample content including the first recommended content; The model obtaining module is configured to obtain the content recognition model; The content recall module is configured to recall second recommended content from the recommended content library by using the content recognition model, the second recommended content being high-quality recommended content identified by the content recognition model, and a recall process of the second recommended content not depending on privacy information of an account; The content recommendation module is configured to determine a permission situation of obtaining privacy information of a target account in a content recommendation process, and perform content recommendation on the target account based on the first recommended content and the second recommended content in a case where the target account closes a permission of obtaining privacy information.

10. A computer device, comprising: The computer device includes a processor and a memory, and the memory stores at least one program, the at least one program is loaded and executed by the processor to implement the content recommendation method of any one of claims 1 to 8.

11. A computer readable storage medium, characterized in that, The readable storage medium stores at least one program, and the at least one program is loaded and executed by the processor to implement the content recommendation method of any one of claims 1 to 8.

12. A computer program product, characterised in that, The computer program product includes computer instructions stored in a computer readable storage medium, and a processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to implement the content recommendation method of any one of claims 1 to 8.

Images