Cold start traffic distribution method and device, electronic equipment and storage medium

Abstract

The application relates to a cold start traffic distribution method and device, electronic equipment and a storage medium. The method comprises the following steps: inputting a content association feature corresponding to target media content in a cold start state into a heat gain prediction model, predicting content heat in a state of no cold start traffic distribution and in a state of multiple preset cold start traffic distribution, obtaining first content heat prediction information corresponding to no cold start traffic distribution and second content heat prediction information corresponding to the multiple preset cold start traffic distribution respectively; obtaining heat gain information corresponding to the multiple preset cold start traffic distribution based on the gain of the multiple second content heat prediction information on the basis of the first content heat prediction information; and selecting target heat gain information meeting a heat gain condition from the multiple heat gain information, and distributing the preset cold start traffic corresponding to the target heat gain information to the target media content. According to the technical scheme provided by the application, the distribution accuracy of the cold start traffic can be improved.

Description

Technical Field

[0001] This application relates to the field of Internet application technology, and in particular to a cold start traffic allocation method, device, electronic device and storage medium. Background Technology

[0002] With the development of internet applications, internet-based content publishing has become increasingly diverse. For example, in media content interaction applications, a large number of publishers release media content. In related technologies, these applications allocate a certain amount of traffic (called cold start traffic) to newly released media content (i.e., media content in a cold start state or in the cold start phase) to help increase its exposure during this phase. However, the current allocation of cold start traffic is not precise enough. For example, a fixed allocation or allocating more cold start traffic to media content with high potential for popularity may lead to some newly released media content not receiving sufficient pre-launch buzz during the cold start phase. In contrast, the latter, due to its already high potential for popularity, may not provide significant support from a large amount of cold start traffic, potentially resulting in a waste of cold start traffic. Furthermore, it may prevent media content with lower potential for popularity from receiving sufficient cold start traffic, resulting in a lower overall proportion of newly released media content becoming popular. Summary of the Invention

[0003] This application provides a cold start traffic allocation method, apparatus, electronic device, and storage medium to improve the accuracy of cold start traffic allocation. This allows newly released content to receive appropriate cold start traffic for sufficient pre-heating, thereby increasing the rate at which newly released content becomes popular, while also conserving cold start traffic. The technical solution of this application is as follows:

[0004] According to a first aspect of the embodiments of this application, a cold start flow allocation method is provided, comprising:

[0005] Obtain the content association features corresponding to the target media content that is in a cold start state;

[0006] The content association features are input into the popularity gain prediction model to predict the popularity of the target media content under the conditions of no cold start traffic and multiple preset cold start traffic, respectively, to obtain the first content popularity prediction information corresponding to the no cold start traffic and the second content popularity prediction information corresponding to each of the multiple preset cold start traffic.

[0007] Based on the gain of each of the second content popularity prediction information on the basis of the first content popularity prediction information, the popularity gain information corresponding to each of the various preset cold start flow rates is obtained.

[0008] Target heat gain information that meets the heat gain conditions is selected from multiple heat gain information, and the preset cold start flow corresponding to the target heat gain information is allocated to the target media content from the multiple preset cold start flow.

[0009] According to a second aspect of the embodiments of this application, a cold start flow distribution device is provided, comprising:

[0010] The content association feature acquisition module is used to acquire the content association features corresponding to the target media content in the cold start state;

[0011] The content popularity prediction module is used to input the content association features into the popularity gain prediction model, and predict the content popularity of the target media content under the conditions of no cold start traffic and multiple preset cold start traffic, respectively, to obtain the first content popularity prediction information corresponding to the no cold start traffic and the second content popularity prediction information corresponding to each of the multiple preset cold start traffic.

[0012] The heat gain acquisition module is used to obtain the heat gain information corresponding to each of the multiple preset cold start flows based on the gain of each of the multiple second content heat prediction information on the basis of the first content heat prediction information.

[0013] The cold start flow allocation module is used to filter out the target heat gain information that meets the heat gain conditions from multiple heat gain information, and allocate the preset cold start flow corresponding to the target heat gain information from the multiple preset cold start flows to the target media content.

[0014] In one possible implementation, the popularity gain prediction model includes a feature processing network and multiple popularity prediction networks, each corresponding one-to-one with the absence of cold start traffic and the allocation of multiple preset cold start traffic; the content popularity prediction module includes:

[0015] The target content feature acquisition unit is used to input the content association features into the feature processing network for feature extraction processing to obtain the target content features;

[0016] A popularity prediction unit is used to input the target content features into the plurality of popularity prediction networks to obtain the first content popularity prediction information and a plurality of second content popularity prediction information.

[0017] In one possible implementation, the heat gain condition includes a gain flow rate ratio condition; the cold start flow rate allocation module includes:

[0018] The gain flow ratio determination unit is used to determine the gain flow ratio information of each heat gain information and the preset cold start flow corresponding to each heat gain information;

[0019] The filtering unit is used to select the target gain flow ratio information that meets the gain flow ratio condition from multiple gain flow ratio information;

[0020] The target heat gain information determination unit is used to determine the heat gain information corresponding to the target gain flow ratio information as the target heat gain information;

[0021] The first cold start flow allocation unit is used to allocate the preset cold start flow corresponding to the target heat gain information from the multiple preset cold start flows to the target media content.

[0022] In one possible implementation, the heat gain condition includes a preset gain threshold; the cold start flow distribution module includes:

[0023] The second cold start traffic allocation unit is used to filter out target heat gain information from the at least one heat gain information when there is at least one heat gain information greater than a preset gain threshold, and allocate the preset cold start traffic corresponding to the target heat gain information from the multiple preset cold start traffic to the target media content.

[0024] In one possible implementation, the cold start traffic allocation module is further configured to determine, when multiple heat gain information are all less than the preset gain threshold, the cold start traffic allocation method corresponding to the target media content in the cold start state is to not allocate cold start traffic.

[0025] In one possible implementation, the content association feature acquisition module includes:

[0026] The first feature acquisition unit is used to extract the content attributes of the target media content and the object attributes of the content publishing object to obtain content attribute features and object attribute features.

[0027] The second feature acquisition unit is used to extract embedded features from the content of different media types in the target media content to obtain media content embedded features.

[0028] The content association feature acquisition unit is used to obtain the content association feature based on the content attribute feature, object attribute feature and media content embedding feature.

[0029] In one possible implementation, the apparatus further includes a module for training a preset gain model to obtain the heat gain prediction model, wherein the preset gain model includes a feature extraction network, multiple initial prediction networks, and a cold start flow allocation preference network, and the multiple initial prediction networks correspond one-to-one with the non-allocation of cold start flow and the allocation of multiple preset cold start flow; the apparatus further includes:

[0030] The training sample acquisition module is used to acquire multiple training sample data, which includes multiple sample media content and cold start traffic allocation indication information and popularity tags corresponding to each of the multiple sample media content. The cold start traffic allocation indication information is used to indicate that no cold start traffic is allocated or any of the preset cold start traffic.

[0031] The feature extraction module is used to input the sample content association features corresponding to the target sample media content and the cold start traffic allocation indication information corresponding to the target sample media content into the feature extraction network for feature extraction processing to obtain shared content features; the target sample media content is any of the sample media content;

[0032] The indicator heat prediction module is used to input the shared content features into the target prediction network corresponding to the target cold start traffic in the multiple initial prediction networks, perform heat prediction under the target cold start traffic, and obtain the sample heat prediction information of the target sample media content under the target cold start traffic; the target cold start traffic is the cold start traffic indicated by the cold start traffic allocation indicator information corresponding to the target sample media content.

[0033] The traffic allocation tendency prediction module is used to input the shared content characteristics into the cold start traffic allocation tendency network to perform cold start traffic allocation prediction and obtain allocation prediction information indicating the target cold start traffic.

[0034] The first loss determination module is used to determine the first loss information based on the sample popularity prediction information and the popularity tag corresponding to the target sample media content;

[0035] The second loss determination module is used to determine the second loss information based on the allocation prediction information and the cold start traffic allocation indication information corresponding to the target sample media content;

[0036] The model parameter adjustment module is used to adjust the parameters of the target prediction network, the parameters of the feature extraction network, and the parameters of the cold start traffic allocation tendency network based on the first loss information and the second loss information, until the training iteration convergence condition is met. The feature extraction network and the multiple initial prediction networks corresponding to the training iteration convergence condition are used as the feature processing network and the multiple popularity prediction networks in the popularity gain prediction model.

[0037] In one possible implementation, the training sample acquisition module includes:

[0038] The sample media content acquisition unit is used to randomly acquire multiple sample media contents from the media content in the cold start state;

[0039] The traffic random allocation unit is used to randomly allocate cold start traffic to the multiple sample media content based on the cold start traffic allocation method of not allocating cold start traffic and allocating multiple preset cold start traffic, and generate cold start traffic allocation indication information corresponding to each of the multiple sample media content.

[0040] The content push unit is used to push the multiple sample media content according to the cold start traffic randomly allocated to the multiple sample media content;

[0041] A popularity tag labeling unit is used to collect push feedback information corresponding to each of the multiple sample media contents within a preset time after the push, and to label the multiple sample media contents with popularity tags according to the push feedback information.

[0042] The training sample acquisition unit is used to construct the multiple training sample data based on the correspondence between the multiple sample media content, the cold start traffic allocation indication information, and the popularity tags. Each training sample data includes sample media content with a corresponding relationship, cold start traffic allocation indication information, and popularity tags.

[0043] According to a third aspect of the embodiments of this application, an electronic device is provided, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the method as described in any one of the first aspects above.

[0044] According to a fourth aspect of the present application, a computer-readable storage medium is provided, wherein when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to perform any of the methods described in the first aspect of the present application.

[0045] According to a fifth aspect of the embodiments of this application, a computer program product is provided, including computer instructions that, when executed by a processor, cause a computer to perform the method described in any one of the first aspects of the embodiments of this application.

[0046] The technical solutions provided by the embodiments of this application have at least the following beneficial effects:

[0047] By inputting the content association features of target media content in a cold start state into a popularity gain prediction model, the popularity of the target media content is predicted under two conditions: no cold start traffic and multiple preset cold start traffic. This yields a first content popularity prediction information for no cold start traffic and second content popularity prediction information for each of the multiple preset cold start traffic. Based on the gains of each of the multiple second content popularity prediction information on the first content popularity prediction information, popularity gain information corresponding to each of the multiple preset cold start traffic is obtained. From these multiple popularity gain information, the preset cold start traffic corresponding to the target popularity gain information that meets the popularity gain conditions is selected and allocated to the target media content. This approach, based on the gain value of multiple preset cold start traffic for content popularity, allows for more reasonable allocation of cold start traffic to each target media content in a cold start state. This ensures sufficient preheating of the target media content in the cold start state while effectively conserving cold start traffic, enabling the target media content to achieve better popularity gain during the cold start phase. This improves the accuracy and effectiveness of cold start traffic, thereby increasing the overall rate of media content in a cold start state becoming popular content.

[0048] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0049] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application, and do not constitute an undue limitation of this application.

[0050] Figure 1 This is a schematic diagram illustrating an application environment according to an exemplary embodiment.

[0051] Figure 2 This is a flowchart illustrating a cold start flow allocation method according to an exemplary embodiment.

[0052] Figure 3 This is a schematic diagram of the structure of a heat gain prediction model according to an exemplary embodiment.

[0053] Figure 4 This is a schematic diagram illustrating a cold start flow allocation method according to an exemplary embodiment.

[0054] Figure 5 This is a flowchart illustrating a model training process for obtaining a heat gain prediction model by training a preset gain model, according to an exemplary embodiment.

[0055] Figure 6 This is a schematic diagram illustrating the structure of a preset gain model according to an exemplary embodiment.

[0056] Figure 7 This is a block diagram of a cold start flow distribution device according to an exemplary embodiment.

[0057] Figure 8 This is a block diagram illustrating an electronic device for cold start flow distribution according to an exemplary embodiment.

[0058] Figure 9 This is a block diagram illustrating an electronic device for cold start flow distribution based on an exemplary embodiment. Detailed Implementation

[0059] Various exemplary embodiments, features, and aspects of this application will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0060] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0061] In this application embodiment, 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.

[0062] Furthermore, to better illustrate this application, numerous specific details are provided in the following detailed embodiments. Those skilled in the art should understand that this application can be implemented without certain specific details. In some instances, methods, means, components, and circuits well-known to those skilled in the art have not been described in detail in order to highlight the main points of this application.

[0063] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or computers-controlled machines to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. AI software technology mainly includes computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0064] In recent years, with the research and progress of artificial intelligence technology, artificial intelligence technology has been widely used in many fields. The solutions provided in the embodiments of this application involve technologies such as machine learning / deep learning, which are specifically illustrated through the following embodiments.

[0065] Please see Figure 1 , Figure 1 This diagram illustrates an application system according to an embodiment of this application. The application system can be used in the cold start flow allocation method of this application. Figure 1 As shown, the application system may include at least server 01 and terminal 02.

[0066] In this embodiment, the server 01 can be used for cold start traffic distribution. For example, the server 01 can be a server of a recommendation platform. The server 01 can include an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0067] In this embodiment, the terminal 02 can be used to trigger cold start traffic allocation, display the results of cold start traffic allocation, etc. The terminal 02 may include physical devices such as smartphones, desktop computers, tablets, laptops, smart speakers, digital assistants, augmented reality (AR) / virtual reality (VR) devices, and smart wearable devices. The physical device may also include software running on it, such as applications. In this embodiment, the operating system running on the terminal 02 may include, but is not limited to, Android, iOS, Linux, and Windows.

[0068] In addition, it should be noted that, Figure 1 The example shown is merely one application environment of the cold start flow allocation method provided in this application.

[0069] In the embodiments described in this specification, the terminal 02 and the server 01 can be directly or indirectly connected through wired or wireless communication, and this application does not limit this connection.

[0070] It should be noted that in the specific implementation of this application, user-related data is involved. When the following embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0071] Before introducing the method embodiments provided in this application, a brief introduction will be given on the application scenarios, related terms or nouns that may be involved in the method embodiments of this application, so as to facilitate the understanding of those skilled in the art.

[0072] Cold start traffic: This refers to the supportive traffic distributed to newly published content by a platform (such as a recommendation platform) during the cold start phase. This traffic can refer to positive actions such as exposure. For example, during this cold start phase, the recommendation platform will design dedicated traffic channels and corresponding support strategies for newly published content. Newly published content refers to content in the cold start phase, i.e., content in a cold start state.

[0073] Popular content: also known as trending content, refers to published content whose exposure exceeds a certain threshold, which is relatively high. Popular content accounts for a significant proportion of the total exposure of all content.

[0074] "Out of the slope": When newly published content receives more than a certain threshold of exposure after a period of time, it is called "out of the slope." This can be seen as a more granular classification of popular content. For example, if the threshold is 10,000 and the time is set to 3 days, and newly published content receives more than 10,000 exposures after 3 days, then it is said that the content has "out of the slope" with 10,000 exposures.

[0075] Outgoing content rate: The ratio of the number of newly published content to the total number of newly published content after a period of time is called the outgoing content rate.

[0076] Gain model: also known as uplift model, it models the gain effect of different behaviors and is a causal relationship model. In the embodiments of this specification, different behaviors may include not allocating cold start flow and allocating multiple preset cold start flow (or multiple cold start flow allocation methods).

[0077] Figure 2 This is a flowchart illustrating a cold start flow allocation method according to an exemplary embodiment. For example... Figure 2 As shown, the method may include the following steps.

[0078] In step S201, the content association features corresponding to the target media content in the cold start state are obtained.

[0079] In the embodiments of this specification, the cold start state can refer to being in the cold start phase. The cold start phase can refer to the period from the time the media content is published until the media content reaches a predetermined exposure or a predetermined publication duration. The predetermined exposure and predetermined publication duration can be set by the actual business, and this application does not limit them.

[0080] The target media content can refer to any media content in a cold start state. The media content can be multimedia content or single media content, such as including but not limited to video, text, images, etc. This application does not limit it.

[0081] Accordingly, the content association features corresponding to the target media content can refer to the content-related features that can characterize the target media content. For example, the content association features may include, but are not limited to, features related to content type, content keywords, and content publishing objects (such as content publishers or content publishing accounts), etc., which are not limited in this application.

[0082] In one possible implementation, obtaining the content association features corresponding to the target media content in a cold start state can include: extracting the content attributes of the target media content and the object attributes of the content publishing object to obtain content attribute features and object attribute features; extracting embedded features from the content of different media types in the target media content to obtain media content embedding features; and thus, content association features can be obtained based on the content attribute features, object attribute features, and media content embedding features. Different media types may include, but are not limited to, text, images, and audio. For example, the content attribute features, object attribute features, and media content embedding features can be used as content association features, or they can be concatenated to form the content association features. By including content attribute features, object attribute features, and media content embedding features in the content association features corresponding to the target media content, the expression of the target media content becomes more comprehensive. This allows cold start traffic allocation prediction to be more adapted to media content recommendations, thereby making cold start traffic allocation more accurate.

[0083] In one example, content attribute features can refer to features that characterize content attributes, such as content type, text keywords, and, when the content is video, video duration and resolution. Object attribute features can refer to features that characterize object attributes, such as the content type tags, region, preferred interactive accounts, recent exposure of the published content, and click-through rate of the published content. Media content embedding features can refer to the embedding features of media content, such as the embedding features of different media types within the target media content. The target media content can include text, images, and other media types; correspondingly, media content embedding features can include text embedding features, image embedding features, etc., and this application does not limit these. Exemplarily, embedded feature extraction can be implemented based on a pre-trained embedding feature extraction model, and this application does not limit this either. Optionally, content attribute features, object attribute features, and media content embedding features can all be vector features.

[0084] In step S203, the content association features are input into the popularity gain prediction model to predict the popularity of the target media content under the conditions of no cold start traffic and multiple preset cold start traffic, respectively, to obtain the first content popularity prediction information corresponding to no cold start traffic and the second content popularity prediction information corresponding to each of the multiple preset cold start traffic.

[0085] In the embodiments of this specification, the popularity gain prediction model can be obtained by training a preset gain model using multiple training sample data. Each training sample data includes sample media content and corresponding cold start traffic allocation indication information and popularity tags. The cold start traffic allocation indication information is used to indicate whether no cold start traffic is allocated or any preset cold start traffic, that is, to indicate the cold start traffic allocation method. The popularity tag can be used to mark whether the sample media content is popular content, and can be distinguished by 0 and 1, for example, 1 represents popular content and 0 represents non-popular content. Exemplarily, the preset gain model may include, but is not limited to, S-Learner, T-Learner, DragonNet, DESCN, FlexTENet, S-Net, etc., and this application does not limit it.

[0086] No cold start flow is allocated, which can be considered as allocating zero cold start flow, represented by X0. Multiple preset cold start flows can be different preset cold start flows, which can be pre-set cold start flows. For example, multiple preset cold start flows may include X1, X2, X3, etc., such as... Figure 3As shown. Cold start traffic refers to the traffic that media content receives during the cold start phase, i.e., the support traffic that media content receives during the cold start stage. For example, in a media content recommendation scenario, this traffic can refer to the number of accounts (users) that can recommend and display the media content, that is, the exposure that the media content receives from being recommended.

[0087] In practical applications, content association features can be input into the popularity gain prediction model to predict the popularity of target media content under different conditions: no cold start traffic and multiple preset cold start traffic. This yields first content popularity prediction information corresponding to no cold start traffic and second content popularity prediction information corresponding to each of the multiple preset cold start traffic conditions. For example, the first and second content popularity prediction information can refer to the probability of being predicted as popular content (e.g., trending or trending content). Based on this, the first and second content popularity prediction information can be popularity prediction probabilities, such as first content popularity prediction probability and second content popularity prediction probability. The range of these first and second content popularity prediction probabilities can be 0 to 1. Optionally, the first and second content popularity prediction information can refer to the exposure volume of the predicted popular content. Based on this, the first and second content popularity prediction information can be first content exposure volume and second content exposure volume, respectively.

[0088] In one alternative implementation, such as Figure 3 As shown, the heat gain prediction model can include a feature processing network and multiple heat prediction networks, such as... Figure 3 The heat prediction networks shown are 0 to 3. These multiple heat prediction networks correspond one-to-one with both no cold start flow allocation and allocation of multiple preset cold start flows. In other words, each heat prediction network corresponds to one of the multiple heat prediction networks, with one network corresponding to one cold start flow allocation method. Accordingly, step S203 may include:

[0089] Content association features are input into a feature processing network for feature extraction to obtain target content features; that is, the output of the feature processing network is the target content feature. For example, this target content feature can be a latent vector feature obtained by processing the content association features through the feature processing network.

[0090] Furthermore, the target content features can be input into multiple popularity prediction networks to predict content popularity under conditions of no cold start traffic and with multiple preset cold start traffic, thereby obtaining first content popularity prediction information and multiple second content popularity prediction information. For example, refer to Figure 3The target content features can be input into popularity prediction networks 0 through 3, respectively. Popularity prediction network 0 predicts content popularity without allocating cold start traffic; popularity prediction network 1 predicts content popularity when allocating a preset cold start traffic of X1; popularity prediction network 2 predicts content popularity when allocating a preset cold start traffic of X2; and popularity prediction network 3 predicts content popularity when allocating a preset cold start traffic of X3. After predicting content popularity through various cold start traffic allocation methods, multiple content popularity prediction information can be obtained. Taking the content popularity prediction information as the probability of being predicted as popular content as an example, the first content popularity prediction information and multiple second content popularity prediction information can be obtained as follows: Figure 3 P0 to P3 are shown.

[0091] In step S205, based on the gains of multiple second content popularity prediction information on the basis of the first content popularity prediction information, popularity gain information corresponding to multiple preset cold start flow rates is obtained.

[0092] In the embodiments of this specification, the gain of the second content popularity prediction information based on the first content popularity prediction information can refer to the gain of the second content popularity prediction information compared to the first content popularity prediction information. The popularity gain information corresponding to each of the multiple preset cold start flows can be used to characterize the increase in the probability of popularity that each of the multiple preset cold start flows can obtain compared to no allocated cold start flow; or it can be used to characterize the increase in exposure that each of the multiple preset cold start flows can obtain compared to no allocated cold start flow, for example, the incremental exposure compared to no allocated cold start flow.

[0093] For example, the popularity gain information corresponding to each preset cold start flow can be the difference between the second content popularity prediction information and the first content popularity prediction information corresponding to each preset cold start flow. For instance, the popularity gain information corresponding to preset cold start flow X1 is P1 - P0. Based on a similar processing method, popularity gain information corresponding to various preset cold start flows can be obtained.

[0094] In step S207, target heat gain information that meets the heat gain conditions is selected from multiple heat gain information, and the preset cold start flow corresponding to the target heat gain information is allocated to the target media content from multiple preset cold start flow.

[0095] In the embodiments of this specification, the heat gain conditions may include, but are not limited to, a preset gain threshold, a gain flow ratio condition, and multiple heat gain information items being sorted from largest to smallest and placed in a preset sorting position. As an example, the preset sorting position can be any sorting position within the top N, where N can be greater than 1 and less than or equal to a preset integer, such as 3. This application does not limit this. Exemplarily, the target heat gain information can be higher than any other heat gain information among the multiple heat gain information items, i.e., the target heat gain information can be the highest heat gain information among the multiple heat gain information items, i.e., the case of being sorted from largest to smallest. For the relevant content of the preset gain threshold and gain flow ratio condition, please refer to the corresponding description below, which will not be repeated here.

[0096] Optionally, if there are multiple target heat gain information pieces that meet the heat gain condition, one target heat gain information piece can be randomly selected from these multiple target heat gain information pieces that meet the heat gain condition, and the preset cold start traffic corresponding to the randomly selected target heat gain information piece can be allocated to the target media content. Alternatively, the highest target heat gain information piece can be selected from the multiple target heat gain information pieces that meet the heat gain condition, and the preset cold start traffic corresponding to the highest target heat gain information piece can be allocated to the target media content.

[0097] The preset cold start flow corresponding to the target heat gain information can refer to the preset cold start flow that generates the target heat gain information, which is the preset cold start flow allocated to the second content heat prediction information corresponding to the target heat gain information.

[0098] In practical applications, such as Figure 3 As shown, for example, if the target heat gain information is selected as (P2-P0) from multiple heat gain information, the preset cold start flow corresponding to the target heat gain information can be determined as X2, and the preset cold start flow X2 can be allocated to the target media content.

[0099] Optionally, the preset cold start traffic corresponding to the target popularity gain information allocated above can be used to push (or recommend) the target media content. For example, the preset cold start traffic corresponding to the target popularity gain information can be used as the support traffic for the target media content in the cold start state.

[0100] By inputting the content association features corresponding to target media content in a cold start state into the popularity gain prediction model, the popularity of the target media content is predicted under the conditions of no cold start traffic and multiple preset cold start traffic. This yields the first content popularity prediction information corresponding to no cold start traffic and the second content popularity prediction information corresponding to each of the multiple preset cold start traffic. Based on the gain of each of the multiple second content popularity prediction information on the basis of the first content popularity prediction information, the popularity gain information corresponding to each of the multiple preset cold start traffic is obtained. From the multiple popularity gain information, the preset cold start traffic corresponding to the target popularity gain information that meets the popularity gain conditions is selected and allocated to the target media content. In this way, based on the gain value of multiple preset cold start traffic for content popularity, more reasonable cold start traffic can be allocated to each target media content in a cold start state. This ensures sufficient preheating of the target media content in the cold start state while effectively saving cold start traffic, allowing the target media content to obtain better popularity gain during the cold start phase. This improves the accuracy and effectiveness of cold start traffic, thereby increasing the overall rate of media content in a cold start state becoming popular content.

[0101] Reference Figure 4 In one possible implementation, the heat gain condition may include a preset gain threshold. Correspondingly, the above-mentioned method of filtering target heat gain information that meets the heat gain condition from multiple heat gain information and allocating the preset cold start traffic corresponding to the target heat gain information from multiple preset cold start traffic to the target media content may include: if at least one heat gain information is greater than the preset gain threshold, filtering target heat gain information from at least one heat gain information and allocating the preset cold start traffic corresponding to the target heat gain information from multiple preset cold start traffic to the target media content. In other words, a determination can be made first as to whether multiple heat gain information exceeds the preset gain threshold. Figure 4 The system determines whether "at least one popularity gain information is greater than a preset gain threshold?". If so, cold start traffic is allocated only when at least one popularity gain information is greater than the preset gain threshold. This allows the target popularity gain information to be selected from the aforementioned at least one popularity gain information, and the preset cold start traffic corresponding to the target popularity gain information is allocated to the target media content from among various preset cold start traffic options. By setting the allocation of cold start traffic only when at least one popularity gain information is greater than the preset gain threshold, cold start traffic is allocated only when it is determined that allocating cold start traffic will increase the probability of becoming popular content, thus avoiding waste of cold start traffic and improving the accuracy of cold start traffic allocation.

[0102] Optionally, if there are multiple target heat gain information that meet the preset gain threshold, a target heat gain information can be selected from these multiple target heat gain information that meet the preset gain threshold based on a preset selection method, thereby allocating the preset cold start traffic corresponding to the selected target heat gain information to the target media content. For example, the preset selection method may include, but is not limited to, random selection or selecting the heat gain information with the highest heat gain information, and this application does not limit this.

[0103] Reference Figure 4 In one possible implementation, the method may further include: when multiple heat gain information values ​​are all less than a preset gain threshold, it can be determined that the cold start traffic allocation method for the target media content in a cold start state is no cold start traffic allocation. For example, when multiple preset cold start traffic values ​​corresponding to different heat gain information values ​​are obtained, it can be determined that all of these multiple heat gain information values ​​are less than the preset gain threshold, i.e. Figure 4 The system determines whether "at least one popularity gain information is greater than the preset gain threshold?". If not, meaning multiple popularity gain information are all less than the preset gain threshold, it can be determined that the cold start traffic allocation method for the target media content in a cold start state is not to allocate cold start traffic. In other words, if multiple preset cold start traffic methods cannot increase the probability of becoming popular content, then no cold start traffic will be allocated. This avoids wasting cold start traffic, improves the accuracy of cold start traffic allocation, maintains the popularity effect of media content, and effectively avoids the negative impact on popularity caused by unnecessary cold start traffic intervention.

[0104] In one possible implementation, the aforementioned heat gain condition may include a gain flow ratio condition. Correspondingly, the process of selecting the target heat gain information that satisfies the heat gain condition from multiple heat gain information sources and allocating the preset cold start flow corresponding to the target heat gain information from multiple preset cold start flows to the target media content may include:

[0105] The gain flow ratio information for each heat gain information and its corresponding preset cold start flow is determined. This gain flow ratio information characterizes the ability of a unit cold start flow to improve heat gain within the preset cold start flow, or in other words, it characterizes the heat gain that a unit cold start flow can improve. For example, if the heat gain information is P1-P0, then the gain flow ratio information of P1-P0 and its corresponding preset cold start flow can be: (P1-P0) / (X1-X0). When X0 is an unallocated cold start flow, i.e., X0 = 0, the gain flow ratio information can be: (P1-P0) / X1. Based on similar processing, the gain flow ratio information corresponding to each heat gain information can be obtained.

[0106] Next, a target gain traffic percentage information that meets the gain traffic percentage condition can be selected from multiple gain traffic percentage information. In one example, the gain traffic percentage condition may include, but is not limited to, a gain traffic percentage threshold, or multiple gain traffic percentages sorted as top M, for example, M can be 3, which is not limited in this application. For example, when M is 1, the target gain traffic percentage information may refer to the highest gain traffic percentage information among multiple gain traffic percentage information.

[0107] Furthermore, the heat gain information corresponding to the target gain traffic ratio information can be determined as the target heat gain information. This allows the preset cold start traffic corresponding to the target heat gain information from a variety of preset cold start traffic sources to be allocated to the target media content. For example, if the obtained target gain traffic ratio information is (P2-P0) / X2, the target heat gain information can be determined to be (P2-P0). Therefore, the preset cold start traffic corresponding to the target heat gain information from a variety of preset cold start traffic sources can be determined to be X2. The preset cold start traffic X2 can then be allocated to the target media content, that is, providing the target media content with preset cold start traffic X2 during the cold start phase when the target media content is in a cold start state.

[0108] For example, X0 = 0, P0 = 0.1; X1 = 100, P1 = 0.2; X2 = 200, P2 = 0.6; X3 = 300, P3 = 0.65. Based on the method for determining the gain traffic ratio information, the gain traffic ratio information can be determined as follows: (0.2-0.1) / 100 = 0.1 / 100, (0.6-0.1) / 200 = 0.25 / 100, (0.65-0.1) / 300 ≈ 0.183 / 100. Comparing these gain traffic ratio information, the largest gain traffic ratio information, i.e., the target gain traffic ratio information, is 0.25 / 100. This preset cold start traffic of X2 is most effective in increasing the probability of content popularity. In other words, under this preset cold start traffic of X2, the cold start traffic consumed per unit increase in content popularity is the lowest. Therefore, it can be determined that the preset cold start traffic X2 should be allocated to the target media content.

[0109] Optionally, when there are multiple gain traffic percentage information pieces that meet the gain traffic percentage condition, one can be randomly selected as the target gain traffic percentage information, thus ensuring a more stable cold start traffic allocated to the target media content. Alternatively, when there are multiple gain traffic percentage information pieces that meet the gain traffic percentage condition, the gain traffic percentage information corresponding to the smallest preset cold start traffic can be selected as the target gain traffic percentage information from among the multiple preset cold start traffic pieces corresponding to these multiple gain traffic percentage information pieces. This ensures sufficient cold start warm-up for the target media content while effectively conserving cold start traffic resources.

[0110] By setting the gain flow ratio information of each popularity gain information and the corresponding preset cold start flow for each popularity gain information, the preset cold start flow can be used to filter and allocate. This can effectively select preset cold start flow with lower cold start flow but greater gain. This can give full play to the role of cold start flow in the cold start flow pool, and can also allocate the saved cold start flow to media content that can better increase popularity gain, thereby increasing the probability of a large number of newly released media content becoming popular content.

[0111] Optionally, after determining the allocated preset cold start traffic, the preset cold start traffic can be used to recommend target media content during the cold start phase. It should be noted that the above content popularity prediction can include predictions of popular content or predictions of content distribution. Compared to traditional manual experience and setting fixed cold start traffic, the embodiments of this specification can achieve personalized allocation of cold start traffic, providing more precise support for cold start traffic allocation. Furthermore, by treating the lack of cold start traffic allocation and various preset cold start traffic as beneficial actions, it is possible to more rationally determine whether to allocate cold start traffic and to allocate preset cold start traffic with better beneficial effects, thus saving cold start traffic and improving the overall distribution rate of newly released content. In addition, through the embodiments of this specification, the saved cold start traffic can be allocated to media content with lower distribution potential, allowing for sufficient cold start pre-heating and distribution. This content is also more likely to become popular, thereby further increasing the incremental distribution of content.

[0112] In one alternative implementation, the popularity gain information can also be used as a reference for recall and ranking during content recommendation. This application does not limit this to any particular implementation, which can further improve the efficiency of media content with high popularity gain information becoming popular content.

[0113] Reference Figure 5 In one possible implementation, the method may further include a model training step of training a preset gain model to obtain the aforementioned heat gain prediction model. This preset gain model may include a feature extraction network, multiple initial prediction networks, and a cold start traffic allocation bias network, such as... Figure 6 As shown. For example, the preset gain model can be DragonNet. Multiple initial prediction networks correspond one-to-one with no cold start traffic and with multiple preset cold start traffic assignments. The model training steps may include:

[0114] In step S501, multiple training sample data are acquired, including multiple sample media content and cold start traffic allocation indication information and popularity tags corresponding to each sample media content.

[0115] In the embodiments of this specification, the sample media content can be published media content. For example, it can be newly published media content, that is, media content in a cold start state. This allows the preset gain model to better learn the impact of media content in a cold start state on the content popularity when no cold start traffic is allocated and when multiple preset cold start traffic are allocated.

[0116] The aforementioned cold start flow allocation indication information can be used to indicate no cold start flow allocation or any preset cold start flow, that is, the cold start flow allocation indication information can be used to distinguish different cold start flow allocation methods. For example, the cold start flow allocation indication information can be an identifier indicating the allocation method of the cold start flow, such as... Figure 6 In this context, X0 indicates that no cold start flow is allocated, and the corresponding cold start flow allocation instruction information can be set to T=0. Based on this, the cold start flow allocation instruction information corresponding to the preset cold start flow X1 can be set to T=1, the cold start flow allocation instruction information corresponding to the preset cold start flow X2 can be set to T=2, and the cold start flow allocation instruction information corresponding to the preset cold start flow X3 can be set to T=3. The numbers 0, 1, 2, and 3 can be seen as identifiers corresponding to the cold start flow allocation method, i.e., cold start flow allocation instruction information.

[0117] The aforementioned popularity tag can refer to a tag that indicates whether media content is popular. For example, the popularity tag can include popular content and non-popular content. For example, popular content can be represented by 1 and non-popular content can be represented by 0. This application does not limit this.

[0118] In step S503, the sample content association features corresponding to the target sample media content and the cold start traffic allocation indication information corresponding to the target sample media content are input into the feature extraction network for feature extraction processing to obtain shared content features.

[0119] In the embodiments of this specification, the target sample media content can be any sample media content. Here, a sample media content is input each time. Based on this, the target sample media content can be regarded as the currently input sample media content.

[0120] For details on the content association features and acquisition methods of the above samples, please refer to the details on the content association features and acquisition methods mentioned above, which will not be repeated here.

[0121] In one possible implementation, the sample content association features corresponding to the target sample media content and the cold start traffic allocation indication information corresponding to the target sample media content can be input into a feature extraction network for feature extraction processing to obtain shared content features. For example, these shared content features can be the output of the feature extraction network, such as latent vector features.

[0122] In step S505, the shared content features are input into the target prediction network corresponding to the target cold start traffic in multiple initial prediction networks to perform popularity prediction under the target cold start traffic, thereby obtaining the sample popularity prediction information of the target sample media content under the target cold start traffic.

[0123] In the embodiments of this specification, the target cold start traffic is the cold start traffic indicated by the cold start traffic allocation instruction information corresponding to the target sample media content. For example... Figure 6 As shown, if the cold start traffic allocation instruction information is 0, then the cold start traffic indicated by the cold start traffic allocation instruction information is X0, that is, the target cold start traffic is X0, and no cold start traffic is allocated. Accordingly, the target prediction network corresponding to the target cold start traffic is the initial prediction network 0. If the cold start traffic allocation instruction information is 1, then the cold start traffic indicated by the cold start traffic allocation instruction information is X1, that is, the preset cold start traffic X1 is allocated. Accordingly, the target prediction network corresponding to the target cold start traffic is the initial prediction network 1.

[0124] Unlike the previous approach where target content features were input into multiple popularity prediction networks, shared content features are input into one of the subsequent initial prediction networks. This initial prediction network corresponds to the cold start traffic allocation instruction information for the target sample media content. In other words, during training, the shared content features of each sample media content undergo only one cold start traffic allocation method, which is indicated by the cold start traffic allocation instruction information corresponding to each sample media content.

[0125] For example, if the cold start traffic allocation indication information corresponding to the target sample media content is 1, then the target cold start traffic is X1, which is the cold start traffic indicated by the cold start traffic allocation indication information corresponding to the target sample media content. This allows us to determine that the target prediction network corresponding to the target cold start traffic among multiple initial prediction networks is initial prediction network 1. Therefore, shared content features can be input into initial prediction network 1 to perform popularity prediction under the preset cold start traffic X1, obtaining the sample popularity prediction information Q1 for the target sample media content under the target cold start traffic. The sample popularity prediction information can refer to the probability of being predicted as popular content (e.g., trending content or content that has gone viral). Optionally, if the indication is one of initial prediction network 0, initial prediction network 2, and initial prediction network 3, then sample popularity prediction information Q0, Q2, or Q3 can be obtained. Optionally, the sample popularity prediction information can refer to the predicted exposure, for example, it can be called the sample predicted exposure.

[0126] In step S507, the shared content features are input into the cold start traffic allocation tendency network to perform cold start traffic allocation prediction, thereby obtaining the allocation prediction information of the target cold start traffic.

[0127] In the embodiments of this specification, the cold start traffic allocation preference network can learn the tendency of target sample media content to be allocated target cold start traffic, and can be used to predict the probability that target sample media content tends to use target cold start traffic. Based on this, the allocation prediction information g can characterize the probability of being allocated target cold start traffic.

[0128] In one possible implementation, the shared content features can be input into a cold start traffic allocation tendency network to predict cold start traffic allocation, that is, to make a tendency prediction of the allocation target cold start traffic and output allocation prediction information indicating the target cold start traffic.

[0129] In step S509, the first loss information is determined based on the sample popularity prediction information and the popularity tag corresponding to the target sample media content.

[0130] In one possible implementation, as described above, the popularity label can be 0 or 1, and correspondingly, the sample popularity prediction information can be the sample popularity prediction probability, which can take the value of 0 to 1. Based on this, a preset loss function can be used to calculate the first loss information between the sample popularity prediction information and the popularity label corresponding to the target sample media content. This preset loss function can be a cross-entropy loss function, etc., and this application does not limit it to this.

[0131] For example, the cross-entropy loss function can be L1=L(y,y^)=ylog(y^)+(1-y)log(1-y^), where y is the popularity label, y^ is the sample popularity prediction probability, and L1 represents the first loss information.

[0132] In another possible implementation, the popularity tag can be the labeled exposure volume, and the sample popularity prediction information can be the sample predicted exposure volume. Based on this, a preset loss function can be used to calculate the first loss information between the sample popularity prediction information and the popularity tag corresponding to the target sample media content. This preset loss function can be the Mean Squared Error (MSE) loss function. Here, a loss calculation is performed for each sample media content. Based on this, the formula for calculating the first loss information using the MSE loss function can be: (y^-y) 2 , where y is the labeled exposure and y^ is the sample predicted exposure.

[0133] Optionally, when the popularity label is the labeled exposure volume and the sample popularity prediction information is the sample predicted exposure volume, corresponding processing can be performed, such as normalization processing, to improve training efficiency.

[0134] In step S511, the second loss information is determined based on the allocation prediction information and the cold start flow allocation instruction information corresponding to the target sample media content.

[0135] In the embodiments of this specification, the allocation prediction information can be the probability that the cold start traffic corresponding to the target sample media content will be allocated indication information, that is, the score of the cold start traffic. For example, the value of this allocation prediction information can be a range consisting of multiple cold start traffic allocation indication information, such as... Figure 6 0~3 shown.

[0136] In one possible implementation, the difference between the allocation prediction information and the cold start traffic allocation indication information corresponding to the target sample media content, or the absolute value of the difference, can be used as the second loss information.

[0137] In step S513, based on the first loss information and the second loss information, the parameters of the target prediction network, the parameters of the feature extraction network, and the parameters of the cold start flow allocation tendency network are adjusted until the training iteration convergence condition is met. The feature extraction network and multiple initial prediction networks corresponding to the training iteration convergence condition are used as the feature processing network and multiple heat prediction networks in the heat gain prediction model.

[0138] In one possible implementation, the total loss information, including the first and second loss information, can be determined. Gradient backpropagation can then be performed based on this total loss information to adjust the parameters of the target prediction network, the feature extraction network, and the cold start traffic allocation bias network until the training iteration convergence condition is met. This allows the feature extraction network and multiple initial prediction networks that meet the training iteration convergence condition to be used as the feature processing network and multiple popularity prediction networks in the popularity gain prediction model. For example, if the target prediction network is initial prediction network 1, then the parameters of initial prediction network 1, the parameter extraction network, and the cold start traffic allocation bias network can be adjusted.

[0139] For example, the convergence condition for training iteration may include, but is not limited to, the total loss information changing for a preset number of times continuously less than the loss threshold, that is, the total loss information remaining basically unchanged for a preset number of times and tending to stabilize. This application does not limit this.

[0140] Optionally, if the current total loss information does not meet the training iteration convergence condition, the next target sample media content in the training sample data can be used to continue the above training process on the preset gain model until the training iteration convergence condition is met. As an example, the selection method for target sample media content can include, but is not limited to, polling, traversal, etc. Alternatively, it can be divided into a training set and a test set. In this way, target sample media content can be selected sequentially from the training set, and when the total loss information is less than the preset loss, the sample media content in the test set can be used for further iterative learning. When the change in the total loss information continues for a preset number of times less than the loss threshold, it is determined that the training iteration convergence condition is met. Thus, the feature extraction network and multiple initial prediction networks corresponding to the satisfaction of the training iteration convergence condition can be used as the feature processing network and multiple popularity prediction networks in the popularity gain prediction model.

[0141] In an optional implementation, step S501 may include the following steps:

[0142] Randomly select multiple sample media contents from media content that is in a cold start state;

[0143] Based on the cold start traffic allocation methods of not allocating cold start traffic and allocating multiple preset cold start traffic, cold start traffic is randomly allocated to multiple sample media content, and cold start traffic allocation indication information corresponding to each sample media content is generated. That is, cold start traffic can be randomly not allocated to each sample media content or any preset cold start traffic can be allocated, and the corresponding cold start traffic allocation indication information is recorded. For example, if no cold start traffic is allocated to sample media content 1, the cold start traffic allocation indication information for sample media content 1 can be recorded as T=0; or, if preset cold start traffic X2 is allocated to sample media content 2, the cold start traffic allocation indication information for sample media content 2 can be recorded as T=2.

[0144] Furthermore, based on the cold start traffic randomly allocated to multiple sample media content as described above, multiple sample media content can be pushed, that is, multiple sample media content can be recommended using the cold start traffic randomly allocated to multiple sample media content as described above.

[0145] Next, the push feedback information corresponding to each of the multiple sample media contents within a preset time period after the push can be statistically analyzed, and popularity tags can be assigned to each of the multiple sample media contents based on the push feedback information. This application does not limit the preset time period; for example, it can be 3 days. Push feedback information can refer to display feedback, interaction feedback, etc., of the sample media content. For example, display feedback can include, but is not limited to, exposure volume; interaction feedback can include, but is not limited to, click-through rate, share rate, etc. Furthermore, popularity tags can be assigned to each sample media content based on the push feedback information corresponding to each sample media content. For example, if the exposure volume reaches the exposure threshold and the click-through rate reaches the click-through rate threshold, the content can be tagged as popular, for example, represented by 1; if the exposure volume does not reach the exposure threshold or the click-through rate does not reach the click-through rate threshold, the content can be tagged as non-popular, for example, represented by 0. This application does not limit this.

[0146] Finally, multiple training sample datasets can be constructed based on the correspondence between multiple sample media content, cold start traffic allocation instructions, and popularity tags. Each training sample dataset can include corresponding sample media content, cold start traffic allocation instructions, and popularity tags. For example, a training sample dataset can be as follows: Sample media content 2, cold start traffic allocation instructions T=2, and popularity tag 1.

[0147] By randomly selecting sample media content and randomly allocating cold start traffic, the causal relationship between sample popularity prediction information and cold start traffic allocation can be made closer. Thus, the effect of cold start traffic allocation can be effectively reflected through sample popularity prediction information, and the training process can accurately learn the impact of cold start traffic allocation on popularity prediction.

[0148] Figure 7 This is a block diagram illustrating a cold start flow distribution device according to an exemplary embodiment. (Refer to...) Figure 7 The device may include:

[0149] The content association feature acquisition module 701 is used to acquire the content association features corresponding to the target media content in the cold start state.

[0150] The content popularity prediction module 703 is used to input the content association features into the popularity gain prediction model, and predict the content popularity of the target media content under the conditions of no cold start traffic and multiple preset cold start traffic, respectively, to obtain the first content popularity prediction information corresponding to the no cold start traffic and the second content popularity prediction information corresponding to each of the multiple preset cold start traffic.

[0151] The heat gain acquisition module 705 is used to obtain the heat gain information corresponding to each of the multiple preset cold start flows based on the gain of each of the multiple second content heat prediction information on the basis of the first content heat prediction information.

[0152] The cold start flow allocation module 707 is used to filter out the target heat gain information that meets the heat gain conditions from multiple heat gain information, and allocate the preset cold start flow corresponding to the target heat gain information from the multiple preset cold start flows to the target media content.

[0153] By setting options for no cold start traffic and multiple preset cold start traffic, the content association features corresponding to target media content in the cold start state are input into the popularity gain prediction model. The model predicts the popularity of the target media content under both conditions, obtaining first content popularity prediction information for no cold start traffic and second content popularity prediction information for each of the preset cold start traffic. Based on the gains of each of the second content popularity prediction information on the first content popularity prediction information, the popularity gain information corresponding to each of the preset cold start traffic is obtained. From the multiple popularity gain information, the preset cold start traffic corresponding to the target popularity gain information that meets the popularity gain conditions is selected and allocated to the target media content. This approach, based on the gain value of multiple preset cold start traffic for content popularity, allows for more reasonable allocation of cold start traffic to each target media content in the cold start state. This ensures sufficient preheating of the target media content in the cold start state while effectively saving cold start traffic, enabling the target media content to obtain better popularity gain during the cold start phase. This improves the accuracy and effectiveness of cold start traffic, thereby increasing the overall rate of media content in the cold start state becoming popular content.

[0154] In one possible implementation, the popularity gain prediction model includes a feature processing network and multiple popularity prediction networks, each corresponding one-to-one with the absence of cold start traffic and the allocation of multiple preset cold start traffic; the content popularity prediction module 703 may include:

[0155] The target content feature acquisition unit is used to input the content association features into the feature processing network for feature extraction processing to obtain the target content features;

[0156] A popularity prediction unit is used to input the target content features into the plurality of popularity prediction networks to obtain the first content popularity prediction information and a plurality of second content popularity prediction information.

[0157] In one possible implementation, the heat gain condition includes a gain flow rate ratio condition; the cold start flow rate allocation module 707 may include:

[0158] The gain flow ratio determination unit is used to determine the gain flow ratio information of each heat gain information and the preset cold start flow corresponding to each heat gain information;

[0159] The filtering unit is used to select the target gain flow ratio information that meets the gain flow ratio condition from multiple gain flow ratio information;

[0160] The target heat gain information determination unit is used to determine the heat gain information corresponding to the target gain flow ratio information as the target heat gain information;

[0161] The first cold start flow allocation unit is used to allocate the preset cold start flow corresponding to the target heat gain information from the multiple preset cold start flows to the target media content.

[0162] In one possible implementation, the heat gain condition includes a preset gain threshold; the cold start flow distribution module 707 may include:

[0163] The second cold start traffic allocation unit is used to filter out target heat gain information from at least one of the heat gain information when there is at least one heat gain information greater than a preset gain threshold, and allocate the preset cold start traffic corresponding to the target heat gain information from the multiple preset cold start traffic to the target media content.

[0164] In one possible implementation, the cold start traffic allocation module is further configured to determine, when multiple heat gain information are all less than the preset gain threshold, the cold start traffic allocation method corresponding to the target media content in the cold start state is to not allocate cold start traffic.

[0165] In one possible implementation, the content association feature acquisition module 701 may include:

[0166] The first feature acquisition unit is used to extract the content attributes of the target media content and the object attributes of the content publishing object to obtain content attribute features and object attribute features.

[0167] The second feature acquisition unit is used to extract embedded features from the content of different media types in the target media content to obtain media content embedded features.

[0168] The content association feature acquisition unit is used to obtain the content association feature based on the content attribute feature, object attribute feature and media content embedding feature.

[0169] In one possible implementation, the apparatus further includes a module for training a preset gain model to obtain the heat gain prediction model, wherein the preset gain model includes a feature extraction network, multiple initial prediction networks, and a cold start flow allocation preference network, and the multiple initial prediction networks correspond one-to-one with the non-allocation of cold start flow and the allocation of multiple preset cold start flow; the apparatus may further include:

[0170] The training sample acquisition module is used to acquire multiple training sample data, which includes multiple sample media content and cold start traffic allocation indication information and popularity tags corresponding to each of the multiple sample media content. The cold start traffic allocation indication information is used to indicate that no cold start traffic is allocated or any of the preset cold start traffic.

[0171] The feature extraction module is used to input the sample content association features corresponding to the target sample media content and the cold start traffic allocation indication information corresponding to the target sample media content into the feature extraction network for feature extraction processing to obtain shared content features; the target sample media content is any of the sample media content;

[0172] The indicator heat prediction module is used to input the shared content features into the target prediction network corresponding to the target cold start traffic in the multiple initial prediction networks, perform heat prediction under the target cold start traffic, and obtain the sample heat prediction information of the target sample media content under the target cold start traffic; the target cold start traffic is the cold start traffic indicated by the cold start traffic allocation indicator information corresponding to the target sample media content.

[0173] The traffic allocation tendency prediction module is used to input the shared content characteristics into the cold start traffic allocation tendency network to perform cold start traffic allocation prediction and obtain allocation prediction information indicating the target cold start traffic.

[0174] The first loss determination module is used to determine the first loss information based on the sample popularity prediction information and the popularity tag corresponding to the target sample media content;

[0175] The second loss determination module is used to determine the second loss information based on the allocation prediction information and the cold start traffic allocation indication information corresponding to the target sample media content;

[0176] The model parameter adjustment module is used to adjust the parameters of the target prediction network, the parameters of the feature extraction network, and the parameters of the cold start traffic allocation tendency network based on the first loss information and the second loss information, until the training iteration convergence condition is met. The feature extraction network and the multiple initial prediction networks corresponding to the training iteration convergence condition are used as the feature processing network and the multiple popularity prediction networks in the popularity gain prediction model.

[0177] In one possible implementation, the training sample acquisition module may include:

[0178] The sample media content acquisition unit is used to randomly acquire multiple sample media contents from the media content in the cold start state;

[0179] The traffic random allocation unit is used to randomly allocate cold start traffic to the multiple sample media content based on the cold start traffic allocation method of not allocating cold start traffic and allocating multiple preset cold start traffic, and generate cold start traffic allocation indication information corresponding to each of the multiple sample media content.

[0180] The content push unit is used to push the multiple sample media content according to the cold start traffic randomly allocated to the multiple sample media content;

[0181] A popularity tag labeling unit is used to collect push feedback information corresponding to each of the multiple sample media contents within a preset time after the push, and to label the multiple sample media contents with popularity tags according to the push feedback information.

[0182] The training sample acquisition unit is used to construct the multiple training sample data based on the correspondence between the multiple sample media content, the cold start traffic allocation indication information, and the popularity tags. Each training sample data includes sample media content with a corresponding relationship, cold start traffic allocation indication information, and popularity tags.

[0183] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0184] Figure 8 This is a block diagram illustrating an electronic device for cold start flow distribution according to an exemplary embodiment. The electronic device may be a terminal, and its internal structure diagram may be as follows: Figure 8As shown, the electronic device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a cold start flow distribution method. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the device's casing, or an external keyboard, touchpad, or mouse.

[0185] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the electronic device to which the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0186] Figure 9 This is a block diagram of an electronic device for cold start traffic distribution based on an exemplary embodiment. The electronic device may be a server, and its internal structure diagram may be as follows. Figure 9 As shown, this electronic device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with external terminals via a network connection. When executed by the processor, the computer program implements a cold start traffic allocation method.

[0187] Those skilled in the art will understand that Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the electronic device to which the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0188] In an exemplary embodiment, an electronic device is also provided, including: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the cold start flow allocation method as described in the embodiments of this application.

[0189] In an exemplary embodiment, a computer-readable storage medium is also provided, which, when executed by a processor of an electronic device, enables the electronic device to perform the cold start traffic allocation method of the present application embodiments. The computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, or optical data storage device, etc.

[0190] In an exemplary embodiment, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to perform the cold start traffic allocation method in the embodiments of this application.

[0191] 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 non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0192] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0193] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A cold start flow distribution method, characterized in that, The method includes: Obtain the content association features corresponding to the target media content that is in a cold start state; The content association features are input into the popularity gain prediction model to predict the popularity of the target media content under the conditions of no cold start traffic and multiple preset cold start traffic, respectively, to obtain the first content popularity prediction information corresponding to the no cold start traffic and the second content popularity prediction information corresponding to each of the multiple preset cold start traffic. Based on the gain of each of the second content popularity prediction information on the basis of the first content popularity prediction information, the popularity gain information corresponding to each of the various preset cold start flow rates is obtained. Target heat gain information that meets the heat gain conditions is selected from multiple heat gain information, and the preset cold start flow corresponding to the target heat gain information is allocated to the target media content from the multiple preset cold start flow.

2. The method according to claim 1, characterized in that, The popularity gain prediction model includes a feature processing network and multiple popularity prediction networks, each corresponding one-to-one with the absence of allocated cold start traffic and the allocation of multiple preset cold start traffic. The step of inputting the content association features into the popularity gain prediction model to predict the content popularity of the target media content under both the absence of allocated cold start traffic and the allocation of multiple preset cold start traffic conditions yields first content popularity prediction information corresponding to the absence of allocated cold start traffic and second content popularity prediction information corresponding to each of the multiple preset cold start traffic conditions, including: The content-related features are input into the feature processing network for feature extraction to obtain the target content features; The target content features are input into the multiple popularity prediction networks respectively to obtain the first content popularity prediction information and multiple second content popularity prediction information.

3. The method according to claim 1, characterized in that, The heat gain condition includes a gain flow ratio condition; the step of filtering out target heat gain information that meets the heat gain condition from multiple heat gain information, and allocating the preset cold start flow corresponding to the target heat gain information from the multiple preset cold start flow to the target media content includes: Determine the gain flow ratio information of each heat gain information and the preset cold start flow corresponding to each heat gain information; Select the target gain flow ratio information that satisfies the gain flow ratio condition from multiple gain flow ratio information; The heat gain information corresponding to the target gain flow ratio information is determined as the target heat gain information; The preset cold start flow corresponding to the target heat gain information is allocated to the target media content from among the various preset cold start flow rates.

4. The method according to any one of claims 1-3, characterized in that, The heat gain condition includes a preset gain threshold; the step of filtering out target heat gain information that meets the heat gain condition from multiple heat gain information, and allocating the preset cold start traffic corresponding to the target heat gain information from the multiple preset cold start traffic to the target media content includes: If at least one of the heat gain information is greater than the preset gain threshold, the target heat gain information is selected from the at least one heat gain information, and the preset cold start traffic corresponding to the target heat gain information from the multiple preset cold start traffic is allocated to the target media content.

5. The method according to claim 4, characterized in that, The method further includes: If multiple heat gain information values ​​are less than the preset gain threshold, the cold start traffic allocation method for the target media content in the cold start state is determined to be no cold start traffic allocation.

6. The method according to claim 1, characterized in that, The acquisition of content association features corresponding to target media content in a cold start state includes: Extract the content attributes of the target media content and the object attributes of the content publishing object to obtain content attribute features and object attribute features; Embedded features are extracted from different media types within the target media content to obtain media content embedded features; Based on the content attribute features, object attribute features, and media content embedding features, the content association features are obtained.

7. The method according to claim 2, characterized in that, The method further includes a model training step of training a preset gain model to obtain the heat gain prediction model. The preset gain model includes a feature extraction network, multiple initial prediction networks, and a cold start flow allocation tendency network. The multiple initial prediction networks correspond one-to-one with the non-allocation of cold start flow and the allocation of multiple preset cold start flow. The model training steps include: Acquire multiple training sample data, which include multiple sample media content and cold start traffic allocation indication information and popularity tags corresponding to each of the multiple sample media content. The cold start traffic allocation indication information is used to indicate that no cold start traffic is allocated or any of the preset cold start traffic. The sample content association features corresponding to the target sample media content and the cold start traffic allocation indication information corresponding to the target sample media content are input into the feature extraction network for feature extraction processing to obtain shared content features; the target sample media content is any of the sample media content. The shared content features are input into the target prediction network corresponding to the target cold start traffic in the plurality of initial prediction networks to perform popularity prediction under the target cold start traffic, thereby obtaining the sample popularity prediction information of the target sample media content under the target cold start traffic; the target cold start traffic is the cold start traffic indicated by the cold start traffic allocation indication information corresponding to the target sample media content. The shared content features are input into the cold start traffic allocation tendency network to perform cold start traffic allocation prediction, thereby obtaining allocation prediction information indicating the target cold start traffic; The first loss information is determined based on the sample popularity prediction information and the popularity tags corresponding to the target sample media content; The second loss information is determined based on the allocation prediction information and the cold start traffic allocation indication information corresponding to the target sample media content; Based on the first loss information and the second loss information, the parameters of the target prediction network, the parameters of the feature extraction network, and the parameters of the cold start traffic allocation tendency network are adjusted until the training iteration convergence condition is met. The feature extraction network and the plurality of initial prediction networks corresponding to the training iteration convergence condition are used as the feature processing network and the plurality of popularity prediction networks in the popularity gain prediction model.

8. The method according to claim 7, characterized in that, The acquisition of multiple training sample data includes: Randomly select multiple sample media contents from media content that is in a cold start state; Based on the cold start traffic allocation methods of not allocating cold start traffic and allocating multiple preset cold start traffic, cold start traffic is randomly allocated to the multiple sample media content, and cold start traffic allocation instruction information corresponding to each of the multiple sample media content is generated. Based on the cold start traffic randomly allocated to the multiple sample media content, the multiple sample media content is pushed. The push feedback information corresponding to each of the multiple sample media contents within a preset time after the push is calculated, and the multiple sample media contents are labeled with popularity tags according to the push feedback information. Based on the correspondence between the multiple sample media contents, the cold start traffic allocation instruction information, and the popularity tags, the multiple training sample data are constructed. Each training sample data includes sample media contents with corresponding relationships, cold start traffic allocation instruction information, and popularity tags.

9. A cold start flow distribution device, characterized in that, include: The content association feature acquisition module is used to acquire the content association features corresponding to the target media content in the cold start state; The content popularity prediction module is used to input the content association features into the popularity gain prediction model, and predict the content popularity of the target media content under the conditions of no cold start traffic and multiple preset cold start traffic, respectively, to obtain the first content popularity prediction information corresponding to the no cold start traffic and the second content popularity prediction information corresponding to each of the multiple preset cold start traffic. The heat gain acquisition module is used to obtain the heat gain information corresponding to each of the multiple preset cold start flows based on the gain of each of the multiple second content heat prediction information on the basis of the first content heat prediction information. The cold start flow allocation module is used to filter out the target heat gain information that meets the heat gain conditions from multiple heat gain information, and allocate the preset cold start flow corresponding to the target heat gain information from the multiple preset cold start flows to the target media content.

10. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the cold start flow allocation method as described in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device is able to perform the cold start flow allocation method as described in any one of claims 1 to 8.

12. A computer program product, characterized in that, Includes computer instructions, which, when executed by a processor, cause the computer to perform the cold start flow allocation method as described in any one of claims 1 to 8.

Images