Method and device for publishing media resources, storage medium, and electronic device

Abstract

This invention discloses a method, apparatus, storage medium, and electronic device for publishing media resources. The method includes: obtaining a target account feature vector, a target media resource feature vector, and a target virtual resource quantity; performing logical processing on the target account feature vector, target media resource feature vector, and target virtual resource quantity using a target prediction network structure in a target prediction neural network model to obtain a target value for a first competitive parameter of the target media resource; and publishing the target media resource to the target account when the target value of the first competitive parameter of the target media resource satisfies a preset target recall condition. This invention solves the technical problem of limited consideration of factors in related media resource recall methods.

Description

Technical Field

[0001] This invention relates to the field of computers, and more specifically, to a method and apparatus for publishing media resources, a storage medium, and an electronic device. Background Technology

[0002] Among related technologies, deep learning has brought new developments to media resource recommendation.

[0003] For example, in the field of ad recommendation, vector retrieval techniques based on ANN or ALSH are used for ad recall. With ad clicks or conversions as the goal, embedding vectors for users and ads are trained separately. Then, the ad most similar to the user is recalled based on the minimum inner product of these vectors. However, this recall method only considers the user's interest in the ad, ignoring certain ad features (such as ad value). For instance, many high-value ads with low user interest may fail to enter the recall and subsequent ranking stages and cannot be recommended to the user.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This invention provides a method and apparatus for publishing media resources, a storage medium and an electronic device, to at least solve the technical problem that the factors considered in media resource recall methods in related technologies are relatively limited.

[0006] According to one aspect of the present invention, a method for publishing media resources is provided, comprising: obtaining a target account feature vector, a target media resource feature vector, and a target virtual resource quantity, wherein the target media resource feature vector includes a set of media resource features of the target media resource, and the target virtual resource quantity is the number of virtual resources transferred out to publish the target media resource to the target account; performing logical processing on the target account feature vector, the target media resource feature vector, and the target virtual resource quantity through a target prediction network structure in a target prediction neural network model to obtain a target value of a first competition parameter of the target media resource, wherein the target value of the first competition parameter of the target media resource is used to determine whether to select the target media resource as a media resource to be published to the target account from the target media resource set, and the target loss value corresponding to the target prediction neural network model satisfies the pre-defined target loss value. The target loss condition is defined as follows: the target loss value is a loss value determined based on a first loss value and a second loss value. The first loss value is a loss value determined based on the sample account feature vector and the sample media resource feature vector. The second loss value is a loss value determined based on the first sample prediction result and the second sample prediction result. The first sample prediction result is the value of a second competition parameter determined based on the sample account feature vector and the sample media resource feature vector. The second sample prediction result is the value of the first competition parameter determined based on the sample account feature vector, the sample media resource feature vector, and the corresponding number of sample virtual resources. When the target value of the first competition parameter of the target media resource satisfies the preset target recall condition, the target media resource is published to the target account. The target recall condition is used to determine the media resources to be published to the target account from the target media resource set.

[0007] Optionally, the step of logically processing the target account feature vector, the target media resource feature vector, and the number of target virtual resources through the target prediction network structure in the target prediction neural network model to obtain the target value of the first competitive parameter of the target media resource includes: performing an inner product operation on the target account feature vector and the target media resource feature vector through the target prediction neural network model to obtain the estimated click-through rate of the target media resource, wherein the estimated click-through rate is used to represent the estimated probability that the target account will click on the target media resource when the target media resource is published to the target account; and logically processing the estimated click-through rate and the number of target virtual resources through the target prediction neural network model to obtain the target value of the first competitive parameter of the target media resource.

[0008] Optionally, the target prediction neural network model performs the logical processing on the estimated click-through rate and the number of target virtual resources using the following formula to arrive at the target value of the first competitive parameter of the target media resource:

[0009] ,

[0010] in, This indicates the target value of the first competitive parameter for the target media resource. This represents the feature vector of the target account. This represents the feature vector of the target media resource, where the target media resource is the first element in the set of target media resources. Individual media resources This indicates the estimated click-through rate. This indicates the quantity of the target virtual resources. It is a positive integer greater than or equal to 1.

[0011] Optionally, the step of publishing the target media resource to the target account when the target value of the first competitive parameter of the target media resource satisfies the preset target recall condition includes: determining that the target value of the first competitive parameter of the target media resource satisfies the target recall condition when the target value of the first competitive parameter of the target media resource ranks in the top N positions in the target media resource set, and publishing the target media resource to the target account, wherein N is a positive integer greater than or equal to 1, and the value of the first competitive parameter of each media resource in the target media resource set is determined based on the target account feature vector, the media resource feature vector of each media resource, and the virtual resource quantity of each media resource, wherein the virtual resource quantity of each media resource is the number of virtual resources that need to be transferred out to publish each media resource to the target account.

[0012] Optionally, obtaining the target account feature vector and the target media resource feature vector includes: inputting the account information of the target account into the target account tower model structure in the target dual-tower model structure of the target prediction neural network model to obtain the target account feature vector; and inputting the target media resource feature vector into the target media resource tower model structure in the target prediction neural network model to obtain the target media resource feature vector.

[0013] Optionally, the method further includes: training the initial predictive neural network model to be trained using a set of sample account information, a set of sample media resource information, and a set of sample virtual resource quantities until the target loss value corresponding to the initial predictive neural network model meets a preset target loss condition, ending the training, and determining the initial predictive neural network model at the end of training as the target predictive neural network model, wherein the initial account tower model structure in the initial predictive neural network model at the end of training is the target account tower model structure in the target predictive neural network model, and the initial media resource tower model structure in the initial predictive neural network model at the end of training is the target media resource tower model structure in the target predictive neural network model; wherein, during the training process, when the target loss value corresponding to the initial predictive neural network model does not meet the target loss condition, the parameters in at least one of the initial account tower model structure and the initial media resource tower model structure in the initial predictive neural network model are adjusted.

[0014] Optionally, training the initial predictive neural network model to be trained using the sample account information set, the sample media resource information set, and the sample virtual resource quantity set includes: performing the i-th round of training on the initial predictive neural network model through the following steps, where i is a positive integer greater than or equal to 2: inputting the sample account information of the sample accounts used in the i-th round from the sample account information set into the initial account tower model structure obtained after the (i-1)-th round of training to obtain the sample account feature vector determined in the i-th round; inputting the sample media resource information of the sample media resources used in the i-th round from the sample media resource information set into the initial media resource tower model structure obtained after the (i-1)-th round of training to obtain the sample media resource feature vector determined in the i-th round; determining the value of the second competition parameter in the i-th round based on the sample account feature vector and the sample media resource feature vector determined in the i-th round; and determining the i-th round's... A loss value is defined as follows: the actual sample click-through rate used in the i-th round indicates whether the sample account used in the i-th round actually clicks on the sample media resource used in the i-th round after the sample media resource used in the i-th round is published to the sample account used in the i-th round; the value of the first competition parameter in the i-th round is determined based on the feature vector of the sample account and the feature vector of the sample media resource determined in the i-th round, and the number of sample virtual resources used in the i-th round; the value of the second competition parameter in the i-th round is determined based on the value of the second competition parameter in the i-th round and the value of the first competition parameter in the i-th round; the target loss value in the i-th round is determined based on the first loss value and the second loss value in the i-th round; when the target loss value in the i-th round does not meet the target loss condition, the parameters in at least one of the initial account tower model structure and the initial media resource tower model structure trained in the (i-1)-th round are adjusted.

[0015] According to another aspect of the present invention, a media resource publishing apparatus is also provided, comprising: an acquisition module, configured to acquire a target account feature vector of a target account, a target media resource feature vector, and a target virtual resource quantity, wherein the target media resource feature vector includes a set of media resource features of the target media resource, and the target virtual resource quantity is the number of virtual resources transferred out to publish the target media resource to the target account; and a processing module, configured to perform logical processing on the target account feature vector, the target media resource feature vector, and the target virtual resource quantity through a target prediction network structure in a target prediction neural network model to obtain a target value of a first competition parameter of the target media resource, wherein the target value of the first competition parameter of the target media resource is used to determine whether to select the target media resource as a media resource to be published to the target account from the target media resource set, and the target loss corresponding to the target prediction neural network model is... The target loss value satisfies a preset target loss condition. The target loss value is a loss value determined based on a first loss value and a second loss value. The first loss value is a loss value determined based on a sample account feature vector and a sample media resource feature vector. The second loss value is a loss value determined based on a first sample prediction result and a second sample prediction result. The first sample prediction result is the value of a second competitive parameter determined based on the sample account feature vector and the sample media resource feature vector. The second sample prediction result is the value of the first competitive parameter determined based on the sample account feature vector, the sample media resource feature vector, and the corresponding number of sample virtual resources. The publishing module is used to publish the target media resource to the target account when the target value of the first competitive parameter of the target media resource satisfies a preset target recall condition. The target recall condition is used to determine the media resource to be published to the target account from the target media resource set.

[0016] According to another aspect of the present invention, a computer-readable storage medium is also provided, wherein a computer program is stored in the computer program, wherein the computer program is configured to execute the above-described media resource publishing method when it is run.

[0017] According to another aspect of the embodiments of this application, a computer program product or computer program is provided, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the media resource publishing method described above.

[0018] According to another aspect of the present invention, an electronic device is also provided, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to execute the above-described media resource publishing method through the computer program.

[0019] In this embodiment of the invention, the target prediction network structure in the target prediction neural network model performs logical processing on the target account feature vector, the target media resource feature vector, and the number of virtual resources to be transferred out to publish the target media resource to the target account, to obtain the target value of the first competitive parameter of the target media resource. When the target value of the first competitive parameter of the target media resource meets the preset target recall condition, the target media resource is published to the target account. Since the number of virtual resources to be transferred out to publish the target media resource to the target account is taken into account, the purpose of increasing the factors considered in the media resource recall method is achieved, thereby solving the technical problem that the factors considered in the media resource recall method in the related art are relatively limited.

[0020] Furthermore, in this embodiment of the invention, during the training process, the target prediction neural network model combines a first loss value determined by the sample account feature vector and the sample media resource feature vector, and a second loss value determined by the first sample prediction result and the second sample prediction result. The first sample prediction result is the value of the second competition parameter determined based on the sample account feature vector and the sample media resource feature vector, and the second sample prediction result is the value of the first competition parameter determined based on the sample account feature vector, the sample media resource feature vector, and the corresponding number of sample virtual resources. In other words, during training, the first and second loss values ​​are combined, and knowledge distillation is used to guide the model training, thereby improving the model's accuracy. Attached Figure Description

[0021] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0022] Figure 1 This is a schematic diagram of an application environment for an optional media resource publishing method according to an embodiment of the present invention;

[0023] Figure 2 This is a flowchart illustrating an optional method for publishing media resources according to an embodiment of this application;

[0024] Figure 3 This is a schematic diagram of the structure of an optional target prediction neural network model according to an embodiment of this application;

[0025] Figure 4 This is a schematic diagram illustrating the training of an optional target prediction neural network model according to an embodiment of this application;

[0026] Figure 5 This is a structural block diagram of an optional media resource publishing device according to an embodiment of this application;

[0027] Figure 6 This is a schematic diagram of the structure of an optional electronic device according to an embodiment of this application;

[0028] Figure 7 This is a structural block diagram of a computer system for an optional electronic device according to an embodiment of this application. Detailed Implementation

[0029] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0030] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0031] Optionally, the key terms involved in the embodiments of the present invention are explained below:

[0032] Recall / Retrieval: In an advertising playback system, selecting a set of ads that match the current user's interests from a massive ad library is called recall, also known as retrieval.

[0033] Coarse sorting: In an advertising playback system, the process of initially sorting a large number of recalled advertisements and selecting the top N advertisements (N can be set arbitrarily, such as 100, 120, 150, etc.) is carried out.

[0034] Precise ranking: In an advertising playback system, the process of precisely ranking the N advertisements selected by rough ranking and selecting the top M (M < N, such as 2, 3, 5, etc.) advertisements.

[0035] Embedding: A vector of fixed length.

[0036] Sample selection bias: The distribution of samples for model training and prediction is different, resulting in estimation bias.

[0037] Item: In the advertising scenario, an item refers to an advertisement, and in the e-commerce scenario, an item refers to a commodity.

[0038] : effective Cost per Mille The total deduction for every thousand exposures.

[0039] Bid: The quantity of virtual resources to be transferred out for publishing the target media resource to the target account. For example, the advertising bid.

[0040] : The target account feature vector.

[0041] : The target media resource feature vector, such as the advertisement feature vector.

[0042] According to one aspect of the embodiments of the present invention, a method for publishing media resources is provided. Optionally, as an alternative implementation, the above method for publishing media resources can be but is not limited to being applied to an application environment as shown in Figure 1 The above application environment includes a user device 102, a network 110, and a server 112. Among them, the user device 102 includes a memory 104, a processor 106, and a display 108. Among them, the above memory is used to store data, including but not limited to the above target account features. The above processor is used to process data, including but not limited to processing the request data of the target account. The above display includes but is not limited to being used to display the target media resource published to the target account.

[0043] Optionally, in this embodiment, the above user device may be a terminal device configured with a target client, and may include but is not limited to at least one of the following: mobile phone (such as Android mobile phone, iOS mobile phone, etc.), laptop computer, tablet computer, handheld computer, MID (Mobile Internet Devices, mobile Internet device), PAD, desktop computer, smart TV, etc. The target client may be a video client, an instant messaging client, a browser client, a game client, an audio client, a shopping client, etc. The above target media resource can be published to the user on the above target client.

[0044] Optionally, the aforementioned network 110 may include, but is not limited to, wired networks and wireless networks, wherein the wired network includes local area networks, metropolitan area networks and wide area networks, and the wireless network includes Bluetooth, WIFI and other networks that enable wireless communication.

[0045] The aforementioned server 112 includes a database 114 and a processing engine 116. The database 114 stores data, including but not limited to target account feature vectors, target media resource feature vectors, and the number of target virtual resources. The processing engine processes the data, including but not limited to performing the following steps:

[0046] Step S102: Logically process the target account feature vector, the target media resource feature vector, and the target virtual resource quantity through the target prediction network structure in the target prediction neural network model to obtain the target value of the first competitive parameter of the target media resource.

[0047] Step S104: If the target value of the first competitive parameter of the target media resource meets the preset target recall condition, the target media resource is published to the target account.

[0048] Alternatively, as an alternative implementation method, such as Figure 2 As shown, the methods for publishing the aforementioned media resources include:

[0049] Step S202: Obtain the target account feature vector, the target media resource feature vector, and the target virtual resource quantity of the target account. The target media resource feature vector includes a set of media resource features of the target media resource, and the target virtual resource quantity is the number of virtual resources that need to be transferred out to publish the target media resource to the target account.

[0050] The target account mentioned above refers to the account used by the target user to log in to the target client. The target account feature vector is the feature vector of the target account, representing the characteristics of the target account in vector form. The characteristics of the target account include, but are not limited to, basic user profile features, historical behavior statistics features, user behavior sequence features, and user behavior interest mining features.

[0051] Basic user profile features include, but are not limited to: user age, gender, province of residence, consumption status, marital status, and education level. Historical user behavior statistics include, but are not limited to: the number of ad clicks in the past month, the number of ad detail page clicks in the past month, the number of video clicks and follows in the past month, the number of "not interested" clicks in the past month, the average number of ad impressions in the past month; and statistical information for the past three months and six months or more. User behavior sequence features include, but are not limited to: information on recently viewed, clicked, and converted ads and client applications.

[0052] The target media resource feature vector is a vector representation of the target media resource's characteristics. The target media resource can be an advertisement. The features of the target media resource include, but are not limited to, context-based features and ad-based features. Context-based features include, but are not limited to, ad placement information (such as ad placement identifiers, ad creative specifications, etc.), device information (device operating system, device network type), and ad placement context information. Ad-based features include, but are not limited to, ad identifiers, creative identifiers, product identifiers, advertiser identifiers, ad categories, creative content keywords, ad keywords, and other features.

[0053] The target virtual resource quantity represents the revenue the advertising platform receives after the platform publishes the advertisement to the target account, including but not limited to the fees paid by the advertiser to the platform.

[0054] Step S204: The target account feature vector, the target media resource feature vector, and the number of target virtual resources are logically processed through the target prediction network structure in the target prediction neural network model to obtain the target value of the first competition parameter of the target media resource. The target value of the first competition parameter of the target media resource is used to determine whether to select the target media resource in the target media resource set as a media resource to be published to the target account. The target loss value corresponding to the target prediction neural network model satisfies a preset target loss condition. The target loss value is a loss value determined based on a first loss value and a second loss value. The first loss value is a loss value determined based on the sample account feature vector and the sample media resource feature vector. The second loss value is a loss value determined based on the first sample prediction result and the second sample prediction result. The first sample prediction result is the value of the second competition parameter determined based on the sample account feature vector and the sample media resource feature vector. The second sample prediction result is the value of the first competition parameter determined based on the sample account feature vector, the sample media resource feature vector, and the corresponding number of sample virtual resources.

[0055] The target value of the first competitive parameter mentioned above can be used as follows: This means that the target account feature vector can be used This indicates that the feature vector of the target media resource is used... This indicates that vectorized search (ALSH) can be used to quickly recall ads across all ads during the ad recall phase. The top N ads with the best scores are entered into the recommendation system. The value of N can be set according to the actual situation, such as 1, 3, 5, etc., which can quickly recall ads from all ads. The highest-rated advertisement.

[0056] Step S206: If the target value of the first competitive parameter of the target media resource satisfies the preset target recall condition, the target media resource is published to the target account, wherein the target recall condition is used to determine the media resource to be published to the target account in the target media resource set.

[0057] Among them, the aforementioned preset target recall conditions can be If the target media resource scores Top N, and the target value of the first competitive parameter of the target media resource is Top N, then the target media resource will be published to the target account.

[0058] Optionally, the step of logically processing the target account feature vector, the target media resource feature vector, and the number of target virtual resources through the target prediction network structure in the target prediction neural network model to obtain the target value of the first competitive parameter of the target media resource includes: performing an inner product operation on the target account feature vector and the target media resource feature vector through the target prediction neural network model to obtain the estimated click-through rate of the target media resource, wherein the estimated click-through rate is used to represent the estimated probability that the target account will click on the target media resource when the target media resource is published to the target account; and logically processing the estimated click-through rate and the number of target virtual resources through the target prediction neural network model to obtain the target value of the first competitive parameter of the target media resource.

[0059] As an optional implementation, the aforementioned estimated click-through rate represents the estimated probability that the target media resource will be clicked by the target account when it is published to the target account. For example, if an advertisement is published to target account A, the estimated probability that target account A will click on the advertisement.

[0060] As an optional implementation, the target account feature vector is as follows: Target media resource feature vector ,pass and The inner product is used to calculate the estimated click-through rate. The calculation formula is as follows:

[0061] ,

[0062] in, It is the dot product of vectors.

[0063] The target value of the first competitive parameter of the target media resource is determined by the following formula:

[0064] ,

[0065] in, This indicates the target value of the first competitive parameter for the target media resource. This represents the feature vector of the target account. This represents the feature vector of the target media resource, where the target media resource is the first element in the set of target media resources. Individual media resources This indicates the estimated click-through rate. This indicates the quantity of the target virtual resources. It is a positive integer greater than or equal to 1.

[0066] ,

[0067] because The value is very small, so the above formula can be further approximated as:

[0068] ,

[0069] Taking the logarithm of both sides, we get:

[0070] ,

[0071] After conversion, the final result will be It is expressed as two vectors and The final score, calculated in the form of an inner product, is:

[0072] ,

[0073] Through vectors , and the number of target virtual resources (For example, ad bids) can be calculated.

[0074] Optionally, the step of publishing the target media resource to the target account when the target value of the first competitive parameter of the target media resource satisfies the preset target recall condition includes: determining that the target value of the first competitive parameter of the target media resource satisfies the target recall condition when the target value of the first competitive parameter of the target media resource ranks in the top N positions in the target media resource set, and publishing the target media resource to the target account, wherein N is a positive integer greater than or equal to 1, and the value of the first competitive parameter of each media resource in the target media resource set is determined based on the target account feature vector, the media resource feature vector of each media resource, and the virtual resource quantity of each media resource, wherein the virtual resource quantity of each media resource is the number of virtual resources that need to be transferred out to publish each media resource to the target account.

[0075] As an optional implementation, the target media resource set may include all advertisements in the ad recall phase, and the value of the first competition parameter for each media resource in the target media resource set can be obtained by the following formula:

[0076] ,

[0077] The value of the first competition parameter for each media resource. For the target account feature vector, The media resource feature vector for each media resource. The number of virtual resources for each media resource (the number of virtual resources transferred out to publish each media resource to the target account). Finally, the ALSH vectorized retrieval algorithm is used for recall. The top N ads by score are entered into the recommendation system. If the target media resource's first competitive parameter is a specific value... If an ad ranks in the top N, the target media resources will be distributed to the target account. This allows for rapid recall across all ads during the ad recall phase using the ALSH vectorized retrieval algorithm. The highest value among the first competitive parameters.

[0078] Optionally, obtaining the target account feature vector and the target media resource feature vector includes: inputting the account information of the target account into the target account tower model structure in the target dual-tower model structure of the target prediction neural network model to obtain the target account feature vector; and inputting the target media resource feature vector into the target media resource tower model structure in the target prediction neural network model to obtain the target media resource feature vector.

[0079] As an optional implementation, such as Figure 3 The target prediction neural network model shown includes a schematic diagram of a target dual-tower model structure. Account information of the target account (including but not limited to gender, age, and address) is input into the target account tower model structure within the target dual-tower model. The target account tower model structure extracts features from the account information to obtain a feature vector for the target account. Media resource information of the target media resource (including but not limited to contextual features and advertising features) is input into the target media resource tower model structure within the target dual-tower model. The target media resource tower model structure extracts features from the media resource information to obtain a target media resource feature vector. The target prediction neural network model performs logical processing on the target account feature vector, the target media resource feature vector, and the number of target virtual resources, outputting the target value of the first competitive parameter for the target media resource.

[0080] Optionally, the method further includes: training the initial predictive neural network model to be trained using a set of sample account information, a set of sample media resource information, and a set of sample virtual resource quantities until the target loss value corresponding to the initial predictive neural network model meets a preset target loss condition, ending the training, and determining the initial predictive neural network model at the end of training as the target predictive neural network model, wherein the initial account tower model structure in the initial predictive neural network model at the end of training is the target account tower model structure in the target predictive neural network model, and the initial media resource tower model structure in the initial predictive neural network model at the end of training is the target media resource tower model structure in the target predictive neural network model; wherein, during the training process, when the target loss value corresponding to the initial predictive neural network model does not meet the target loss condition, the parameters in at least one of the initial account tower model structure and the initial media resource tower model structure in the initial predictive neural network model are adjusted.

[0081] As an optional implementation, the sample account information set includes account information for sample accounts, which may be accounts logged into the target client. The sample media resource information set includes sample information for sample media, including but not limited to exposed samples, clicked samples, and unexposed samples. Exposed samples include sample media that have been ranked and scored but have not been published. The sample virtual resource quantity set represents the number of virtual resources that need to be transferred out to publish sample media resources to sample accounts.

[0082] Ranking is used during the training of the initial predictive neural network model to be trained. The model uses distillation to address the recall. Model (Initial predictive neural network model for recall) Model training, such as Figure 4 The sorting shown Model and Recall A schematic diagram of the model's distillation training structure.

[0083] In sorting Model and Recall Both models include an initial dual-tower model structure (initial account tower model structure and initial media resource tower model structure), the difference being in recall. The model incorporates the number of sample virtual resources as input. The initial account tower model takes sample account information as input and extracts features from this information to obtain sample account feature vectors. Similarly, the initial media resource tower model takes sample media resources as input and extracts features from this information to obtain sample media resource feature vectors.

[0084] The second loss value mentioned above can be obtained using the following loss function:

[0085] ,

[0086] The prediction results for the first sample above, The above are the prediction results for the second sample. First sample prediction results. It is sorting The model's output, It is a recall The output of the model.

[0087] The first loss value mentioned above is for sorting. The value of the second competing parameter output by the model, along with the actual sample click-through rate (the actual sample click-through rate is known when the sample media resource is published to the sample account and whether the sample account clicks on the sample media resource), can be used to calculate the first loss value (loss) using the cross-entropy loss function. The input to the cross-entropy loss function is the ranking. The value of the second competitive parameter output by the model and the actual sample click rate are used to output the first loss value, loss.

[0088] Target loss value L = loss + Distillation loss was added to the loss function during model training. Improve recall Models and sorting The model achieves consistent scoring across the recall space, thereby improving ad recall performance. This is achieved through training to ultimately generate... Recall model (target prediction neural network model).

[0089] Optionally, training the initial predictive neural network model to be trained using the sample account information set, the sample media resource information set, and the sample virtual resource quantity set includes: performing the i-th round of training on the initial predictive neural network model through the following steps, where i is a positive integer greater than or equal to 2: inputting the sample account information of the sample accounts used in the i-th round from the sample account information set into the initial account tower model structure obtained after the (i-1)-th round of training to obtain the sample account feature vector determined in the i-th round; inputting the sample media resource information of the sample media resources used in the i-th round from the sample media resource information set into the initial media resource tower model structure obtained after the (i-1)-th round of training to obtain the sample media resource feature vector determined in the i-th round; determining the value of the second competition parameter in the i-th round based on the sample account feature vector and the sample media resource feature vector determined in the i-th round; and determining the i-th round's... A loss value is defined as follows: the actual sample click-through rate used in the i-th round indicates whether the sample account used in the i-th round actually clicks on the sample media resource used in the i-th round after the sample media resource used in the i-th round is published to the sample account used in the i-th round; the value of the first competition parameter in the i-th round is determined based on the feature vector of the sample account and the feature vector of the sample media resource determined in the i-th round, and the number of sample virtual resources used in the i-th round; the value of the second competition parameter in the i-th round is determined based on the value of the second competition parameter in the i-th round and the value of the first competition parameter in the i-th round; the target loss value in the i-th round is determined based on the first loss value and the second loss value in the i-th round; when the target loss value in the i-th round does not meet the target loss condition, the parameters in at least one of the initial account tower model structure and the initial media resource tower model structure trained in the (i-1)-th round are adjusted.

[0090] As an optional implementation, assume that the sample account information of the sample account used in the i-th round is the account information of sample account S (including but not limited to the gender, age, address, etc. of account S), and the sample media resource information of the sample media resource used in the i-th round is the media resource information of media resource T (including but not limited to advertising identifiers, creative identifiers, product identifiers, advertiser identifiers, advertising categories, creative content keywords, and advertising keywords). Input the account information of sample account S and the media resource information of media resource T into... Figure 4The ranking model shown obtains the sample account feature vector of sample account S (the sample account feature vector determined in the i-th round) and the media resource feature vector of media resource T (the sample media resource feature vector determined in the i-th round) through the initial account tower model structure and the initial media resource tower model structure in the initial dual-tower model structure. The ranking model performs logical processing on the sample account feature vector of sample account S and the media resource feature vector of media resource T to obtain the value of the second competition parameter in the i-th round. Since the sample account S and media resource T are used as training samples, and it is known whether the sample account S clicked on the media resource T (the actual sample click rate used in the i-th round is known), inputting the value of the second competition parameter in the i-th round and the actual sample click rate used in the i-th round into the cross-entropy loss function can obtain the first loss value loss_i in the i-th round.

[0091] Input the account information of sample account S and the media resource information of media resource T into Figure 4 The recall shown The model, through the initial account tower model structure and the initial media resource tower model structure in the initial dual-tower model structure, obtains the sample account feature vector of sample account S (the sample account feature vector determined in the i-th round) and the media resource feature vector of media resource T (the sample media resource feature vector determined in the i-th round). Recall The initial prediction network structure in the model performs logical processing on the sample account feature vector of sample account S, the media resource feature vector of media resource T, and the number of sample virtual resources used in the i-th round (the number of virtual resources that need to be transferred out when media resource T is published to sample account S) to obtain the value of the first competition parameter in the i-th round, and then obtains the second loss value in the i-th round according to the following formula. :

[0092] ,

[0093] Let the value of the second competition parameter in the i-th round be denoted as described above. Let be the value of the first competition parameter in the i-th round.

[0094] The target loss value for round i is L_i = loss_i + If L_i meets the target loss condition (e.g., L_i is less than a preset threshold, which can be set according to actual conditions, such as 0.1, 0.01, etc.), then training stops, and the recall obtained at the time of stopping training is used to calculate the loss. The model is determined to be a target prediction neural network model. If L_i does not meet the target loss condition (e.g., L_i is less than, greater than, or equal to a preset threshold), then the ranking can be adjusted. Model and Recall The model parameters in at least one of the towers in the dual-tower structure of the model are adjusted, and training continues until the target loss condition is met, thus obtaining the target prediction neural network model.

[0095] As an optional implementation, to improve the model's ability to distinguish between non-exposed samples during the sample sampling process, random negative samples within a batch can be used. To reduce implementation costs, n ad vectors can be randomly selected for each clicked positive sample and concatenated with the user vector of that positive sample to form a random negative sample.

[0096] As an optional implementation method, training Prediction Model: The model generates user vectors based on the training samples from ad clicks. and ad vectors Then through and The inner product is used to calculate the estimated click-through rate. The calculation formula is as follows: in For vector dot product, This step is represented as User vectors and ad vectors generated by the model. Vectors can also be used in the recall phase. Model generation.

[0097] In the above embodiments, the vector inner product formula is modified to satisfy any objective. Furthermore, by introducing model distillation and an auxiliary loss function, the problem of sample selection bias in the recall model is effectively mitigated, thereby greatly improving the accuracy of vector retrieval and the advertising recall effect.

[0098] It is understood that in the specific embodiments of this application, data such as user information are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0099] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0100] According to another aspect of the embodiments of this application, a media resource publishing apparatus for implementing the above-described media resource publishing method is also provided. Figure 5 This is a schematic diagram of the structure of an optional media resource publishing device according to an embodiment of this application, such as... Figure 5 As shown, the device may include: an acquisition module 52, used to acquire a target account feature vector, a target media resource feature vector, and a target virtual resource quantity, wherein the target media resource feature vector includes a set of media resource features of the target media resource, and the target virtual resource quantity is the number of virtual resources that need to be transferred to publish the target media resource to the target account; and a processing module 54, used to perform logical processing on the target account feature vector, the target media resource feature vector, and the target virtual resource quantity through the target prediction network structure in the target prediction neural network model to obtain a target value of a first competition parameter of the target media resource, wherein the target value of the first competition parameter of the target media resource is used to determine whether to select the target media resource as a media resource to be published to the target account in the target media resource set, and the target loss value corresponding to the target prediction neural network model satisfies a preset target loss. The target loss value is determined based on a first loss value and a second loss value. The first loss value is determined based on the sample account feature vector and the sample media resource feature vector. The second loss value is determined based on the first sample prediction result and the second sample prediction result. The first sample prediction result is the value of the second competition parameter determined based on the sample account feature vector and the sample media resource feature vector. The second sample prediction result is the value of the first competition parameter determined based on the sample account feature vector, the sample media resource feature vector, and the corresponding number of sample virtual resources. The publishing module 56 is used to publish the target media resource to the target account when the target value of the first competition parameter of the target media resource meets the preset target recall condition. The target recall condition is used to determine the media resource to be published to the target account in the target media resource set.

[0101] Optionally, the above-mentioned device is further configured to perform an inner product operation on the feature vector of the target account and the feature vector of the target media resource through the target prediction neural network model to obtain the estimated click-through rate of the target media resource, wherein the estimated click-through rate is used to represent the estimated probability that the target account will click on the target media resource when the target media resource is published to the target account; and to perform logical processing on the estimated click-through rate and the number of target virtual resources through the target prediction neural network model to obtain the target value of the first competitive parameter of the target media resource.

[0102] Optionally, the target prediction neural network model performs the logical processing on the estimated click-through rate and the number of target virtual resources using the following formula to arrive at the target value of the first competitive parameter of the target media resource:

[0103] ,

[0104] in, This indicates the target value of the first competitive parameter for the target media resource. This represents the feature vector of the target account. This represents the feature vector of the target media resource, where the target media resource is the first element in the set of target media resources. Individual media resources This indicates the estimated click-through rate. This indicates the quantity of the target virtual resources. It is a positive integer greater than or equal to 1.

[0105] Optionally, the above-described apparatus is further configured to, when the target value of the first competitive parameter of the target media resource ranks among the top N in the target media resource set, determine that the target value of the first competitive parameter of the target media resource satisfies the target recall condition, and publish the target media resource to the target account, wherein N is a positive integer greater than or equal to 1, and the value of the first competitive parameter of each media resource in the target media resource set is determined jointly based on the target account feature vector, the media resource feature vector of each media resource, and the virtual resource quantity of each media resource, wherein the virtual resource quantity of each media resource is the number of virtual resources that need to be transferred out to publish each media resource to the target account.

[0106] Optionally, the above-mentioned device is further configured to input the account information of the target account into the target account tower model structure in the target dual-tower model structure of the target prediction neural network model to obtain the target account feature vector; and input the target media resource feature vector into the target media resource tower model structure in the target prediction neural network model to obtain the target media resource feature vector.

[0107] Optionally, the above-mentioned device is further configured to train the initial predictive neural network model to be trained using a sample account information set, a sample media resource information set, and a sample virtual resource quantity set until the target loss value corresponding to the initial predictive neural network model meets a preset target loss condition, thereby ending the training and determining the initial predictive neural network model at the end of training as the target predictive neural network model. Specifically, the initial account tower model structure in the initial predictive neural network model at the end of training is the target account tower model structure in the target predictive neural network model, and the initial media resource tower model structure in the initial predictive neural network model at the end of training is the target media resource tower model structure in the target predictive neural network model. During the training process, if the target loss value corresponding to the initial predictive neural network model does not meet the target loss condition, the parameters in at least one of the initial account tower model structure and the initial media resource tower model structure in the initial predictive neural network model are adjusted.

[0108] Optionally, the above-described apparatus is further configured to perform the following steps to train the initial predictive neural network model in the i-th round, where i is a positive integer greater than or equal to 2: inputting the sample account information of the sample accounts used in the i-th round from the sample account information set into the initial account tower model structure obtained after the (i-1)-th round of training, to obtain the sample account feature vector determined in the i-th round; inputting the sample media resource information of the sample media resources used in the i-th round from the sample media resource information set into the initial media resource tower model structure obtained after the (i-1)-th round of training, to obtain the sample media resource feature vector determined in the i-th round; determining the value of the second competition parameter in the i-th round based on the sample account feature vector and the sample media resource feature vector determined in the i-th round; and determining the first loss value in the i-th round based on the value of the second competition parameter and the actual sample click-through rate used in the i-th round, wherein the actual sample click-through rate used in the i-th round is used to represent the loss value in the i-th round when the sample media resource information set is used to train the initial predictive neural network model in the (i-1)-th round of training, to obtain the sample media resource feature vector determined in the i-th round; determining the value of the second competition parameter and the actual sample click-through rate used in the i-th round, wherein the actual sample click-through rate used in the i-th round is used to represent the loss value in the i-th round when the sample media resource information set is used to train the initial predictive neural network model in the i-th round of training, to obtain the sample media resource feature vector determined ... After the sample media resources used in the i-th round are published to the sample accounts used in the i-th round, whether the sample accounts used in the i-th round actually click on the sample media resources used in the i-th round; based on the feature vector of the sample accounts determined in the i-th round, the feature vector of the sample media resources determined in the i-th round, and the number of sample virtual resources used in the i-th round, the value of the first competition parameter in the i-th round is determined, and based on the value of the second competition parameter in the i-th round and the value of the first competition parameter in the i-th round, the second loss value in the i-th round is determined; based on the first loss value in the i-th round and the second loss value in the i-th round, the target loss value in the i-th round is determined; when the target loss value in the i-th round does not meet the target loss condition, the parameters in at least one of the initial account tower model structure and the initial media resource tower model structure trained in the (i-1)-th round are adjusted.

[0109] According to another aspect of the embodiments of this application, an electronic device for implementing the above-described method for publishing media resources is also provided. This electronic device may be... Figure 1 The terminal device or server shown. This embodiment uses this electronic device as an example for illustration. Figure 6 As shown, the electronic device includes a memory 602 and a processor 604. The memory 602 stores a computer program, and the processor 604 is configured to execute the steps in any of the above method embodiments via the computer program.

[0110] Optionally, in this embodiment, the aforementioned electronic device may be located in at least one of a plurality of network devices in a computer network.

[0111] Optionally, in this embodiment, the processor can be configured to perform the following steps via a computer program:

[0112] S1, obtain the target account feature vector, the target media resource feature vector, and the target virtual resource quantity of the target account, wherein the target media resource feature vector includes a set of media resource features of the target media resource, and the target virtual resource quantity is the number of virtual resources that need to be transferred out to publish the target media resource to the target account;

[0113] S2, the target prediction neural network structure in the target prediction neural network model performs logical processing on the target account feature vector, the target media resource feature vector, and the number of target virtual resources to obtain the target value of the first competition parameter of the target media resource. The target value of the first competition parameter of the target media resource is used to determine whether to select the target media resource in the target media resource set as a media resource to be published to the target account. The target loss value corresponding to the target prediction neural network model satisfies a preset target loss condition. The target loss value is a loss value determined based on a first loss value and a second loss value. The first loss value is a loss value determined based on the sample account feature vector and the sample media resource feature vector. The second loss value is a loss value determined based on the first sample prediction result and the second sample prediction result. The first sample prediction result is the value of the second competition parameter determined based on the sample account feature vector and the sample media resource feature vector. The second sample prediction result is the value of the first competition parameter determined based on the sample account feature vector, the sample media resource feature vector, and the corresponding number of sample virtual resources.

[0114] S3, if the target value of the first competitive parameter of the target media resource satisfies the preset target recall condition, the target media resource is published to the target account, wherein the target recall condition is used to determine the media resource to be published to the target account in the target media resource set.

[0115] Alternatively, as those skilled in the art will understand, Figure 6 The structure shown is for illustrative purposes only. Electronic devices can also be smartphones (such as Android phones, iOS phones, etc.), tablets, PDAs, and other terminal devices such as MIDs and PADs. Figure 6 This does not limit the structure of the aforementioned electronic devices. For example, the electronic device may also include components that are more... Figure 6 The more or fewer components shown (such as network interfaces, etc.), or having the same Figure 6 The different configurations shown.

[0116] The memory 602 can be used to store software programs and modules, such as the program instructions / modules corresponding to the media resource publishing method and apparatus in this embodiment. The processor 604 executes various functional applications and data processing by running the software programs and modules stored in the memory 602, thereby realizing the aforementioned media resource publishing method. The memory 602 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 602 may further include memory remotely located relative to the processor 604, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof. Specifically, the memory 602 may be used, but is not limited to, for serializing files and compiling files, etc. As an example, such as... Figure 6 As shown, the memory 602 may include, but is not limited to, the acquisition module 52, processing module 54, and publishing module 56 of the media resource publishing device. Furthermore, it may include, but is not limited to, other module units of the media resource publishing device, which will not be elaborated upon in this example.

[0117] Optionally, the transmission device 606 described above is used to receive or send data via a network. Specific examples of the network described above may include wired networks and wireless networks. In one example, the transmission device 606 includes a Network Interface Controller (NIC), which can be connected to other network devices and a router via a network cable to communicate with the Internet or a local area network. In another example, the transmission device 606 is a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0118] In addition, the aforementioned electronic device also includes: a display 608 for displaying media resources; and a connection bus 610 for connecting the various module components in the aforementioned electronic device.

[0119] In other embodiments, the aforementioned terminal device or server can be a node in a distributed system, wherein the distributed system can be a blockchain system, which is a distributed system formed by connecting multiple nodes through network communication. The nodes can form a peer-to-peer (P2P) network, and any form of computing device, such as a server, terminal, or other electronic device, can become a node in the blockchain system by joining this peer-to-peer network.

[0120] According to one aspect of this application, a computer program product is provided, comprising a computer program / instructions containing program code for performing the methods shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via communication section 709, and / or installed from removable medium 711. When the computer program is executed by central processing unit 701, it performs various functions provided in the embodiments of this application. The above embodiment numbers are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0121] Figure 7 A schematic block diagram of a computer system architecture for implementing embodiments of the present application is shown. Figure 7 As shown, the computer system 700 includes a central processing unit (CPU) 701, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 702 or programs loaded from storage section 708 into random access memory (RAM) 703. The RAM 703 also stores various programs and data required for system operation. The CPU 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output interface 705 (I / O interface) is also connected to the bus 704.

[0122] The following components are connected to the input / output interface 705: an input section 706 including a keyboard, mouse, etc.; an output section 707 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a local area network card, modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the input / output interface 705 as needed. A removable medium 711, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 710 as needed so that computer programs read from it can be installed into the storage section 708 as needed.

[0123] Specifically, according to embodiments of this application, the processes described in the various method flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 709, and / or installed from removable medium 711. When the computer program is executed by central processing unit 701, it performs various functions defined in the system of this application.

[0124] It should be noted that, Figure 7 The computer system 700 of the electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0125] According to one aspect of this application, a computer-readable storage medium is provided, from which a processor of a computer device reads computer instructions, and the processor executes the computer instructions, causing the computer device to perform the methods provided in various optional implementations of the above embodiments.

[0126] Optionally, in this embodiment, the computer-readable storage medium may be configured to store a computer program for performing the following steps:

[0127] S1, retrieve the target account feature vector, the target media resource feature vector, and the target virtual resource quantity of the target account, wherein the target media resource feature vector includes a set of media resource features of the target media resource, and the target virtual resource quantity is the number of virtual resources that need to be transferred out to publish the target media resource to the target account;

[0128] S2, the target prediction neural network structure in the target prediction neural network model performs logical processing on the target account feature vector, the target media resource feature vector, and the number of target virtual resources to obtain the target value of the first competition parameter of the target media resource. The target value of the first competition parameter of the target media resource is used to determine whether to select the target media resource in the target media resource set as a media resource to be published to the target account. The target loss value corresponding to the target prediction neural network model satisfies a preset target loss condition. The target loss value is a loss value determined based on a first loss value and a second loss value. The first loss value is a loss value determined based on the sample account feature vector and the sample media resource feature vector. The second loss value is a loss value determined based on the first sample prediction result and the second sample prediction result. The first sample prediction result is the value of the second competition parameter determined based on the sample account feature vector and the sample media resource feature vector. The second sample prediction result is the value of the first competition parameter determined based on the sample account feature vector, the sample media resource feature vector, and the corresponding number of sample virtual resources.

[0129] S3, if the target value of the first competitive parameter of the target media resource satisfies the preset target recall condition, the target media resource is published to the target account, wherein the target recall condition is used to determine the media resource to be published to the target account in the target media resource set.

[0130] Optionally, in this embodiment, those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0131] If the integrated units in the above embodiments are implemented as software functional units and sold or used as independent products, they can be stored in the aforementioned computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause one or more computer devices (which may be personal computers, servers, or network devices, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0132] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0133] In the several embodiments provided in this application, it should be understood that the disclosed client can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection between units or modules, and may be electrical or other forms.

[0134] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0135] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or at least two units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0136] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for publishing media resources, characterized in that, include: Obtain the target account feature vector, the target media resource feature vector, and the target virtual resource quantity of the target account. The target media resource feature vector includes a set of media resource features of the target media resource, and the target virtual resource quantity is the number of virtual resources that need to be transferred out to publish the target media resource to the target account. The target prediction neural network model logically processes the target account feature vector, the target media resource feature vector, and the number of target virtual resources to obtain the target value of the first competition parameter of the target media resource. The target value of the first competition parameter of the target media resource is used to determine whether to select the target media resource from the target media resource set as a media resource to be published to the target account. The target loss value corresponding to the target prediction neural network model satisfies a preset target loss condition. The target loss value is a loss value determined based on a first loss value and a second loss value. The first loss value is a loss value determined based on the sample account feature vector and the sample media resource feature vector. The second loss value is a loss value determined based on the first sample prediction result and the second sample prediction result. The first sample prediction result is the value of the second competition parameter determined based on the sample account feature vector and the sample media resource feature vector. The second sample prediction result is the value of the first competition parameter determined based on the sample account feature vector, the sample media resource feature vector, and the corresponding number of sample virtual resources. If the target value of the first competitive parameter of the target media resource satisfies the preset target recall condition, the target media resource is published to the target account, wherein the target recall condition is used to determine the media resource to be published to the target account in the target media resource set.

2. The method according to claim 1, characterized in that, The step of logically processing the target account feature vector, the target media resource feature vector, and the target virtual resource quantity through the target prediction network structure in the target prediction neural network model to obtain the target value of the first competition parameter of the target media resource includes: The target prediction neural network model performs an inner product operation on the feature vector of the target account and the feature vector of the target media resource to obtain the estimated click-through rate of the target media resource. The estimated click-through rate is used to represent the estimated probability that the target account will click on the target media resource when the target media resource is published to the target account. The target prediction neural network model performs logical processing on the estimated click-through rate and the number of target virtual resources to obtain the target value of the first competitive parameter of the target media resource.

3. The method according to claim 2, characterized in that, The target prediction neural network model performs the logical processing on the estimated click-through rate and the number of target virtual resources using the following formula to arrive at the target value of the first competitive parameter of the target media resource: , in, This indicates the target value of the first competitive parameter for the target media resource. This represents the feature vector of the target account. This represents the feature vector of the target media resource, where the target media resource is the first element in the set of target media resources. Individual media resources This indicates the estimated click-through rate. This indicates the quantity of the target virtual resources. It is a positive integer greater than or equal to 1.

4. The method according to claim 1, characterized in that, The step of publishing the target media resource to the target account when the target value of the first competitive parameter of the target media resource meets the preset target recall condition includes: If the target value of the first competitive parameter of the target media resource ranks among the top N in the target media resource set, it is determined that the target value of the first competitive parameter of the target media resource satisfies the target recall condition, and the target media resource is published to the target account. Here, N is a positive integer greater than or equal to 1. The value of the first competitive parameter of each media resource in the target media resource set is determined based on the target account feature vector, the media resource feature vector of each media resource, and the virtual resource quantity of each media resource. The virtual resource quantity of each media resource is the number of virtual resources that need to be transferred out to publish each media resource to the target account.

5. The method according to claim 1, characterized in that, The acquisition of the target account feature vector and the target media resource feature vector includes: The account information of the target account is input into the target account tower model structure in the target double tower model structure of the target prediction neural network model to obtain the feature vector of the target account; The target media resource feature vector is input into the target media resource tower model structure in the target prediction neural network model to obtain the target media resource feature vector.

6. The method according to claim 1, characterized in that, The method further includes: The initial predictive neural network model to be trained is trained using a set of sample account information, a set of sample media resource information, and a set of sample virtual resource quantities until the target loss value corresponding to the initial predictive neural network model meets the preset target loss condition. The training ends, and the initial predictive neural network model at the end of training is determined as the target predictive neural network model. The initial account tower model structure in the initial predictive neural network model at the end of training is the target account tower model structure in the target predictive neural network model, and the initial media resource tower model structure in the initial predictive neural network model at the end of training is the target media resource tower model structure in the target predictive neural network model. During the training process, when the target loss value corresponding to the initial prediction neural network model does not meet the target loss condition, the parameters in at least one of the initial account tower model structure and the initial media resource tower model structure in the initial prediction neural network model are adjusted.

7. The method according to claim 6, characterized in that, The process of training the initial predictive neural network model to be trained using the sample account information set, sample media resource information set, and sample virtual resource quantity set includes: The initial predictive neural network model is trained in the i-th round using the following steps, where i is a positive integer greater than or equal to 2: Input the sample account information of the sample account used in the i-th round in the sample account information set into the initial account tower model structure obtained after training in the (i-1)-th round to obtain the sample account feature vector determined in the i-th round; and input the sample media resource information of the sample media resources used in the i-th round in the sample media resource information set into the initial media resource tower model structure obtained after training in the (i-1)-th round to obtain the sample media resource feature vector determined in the i-th round. Based on the sample account feature vector determined in the i-th round and the sample media resource feature vector determined in the i-th round, the value of the second competition parameter in the i-th round is determined, and based on the value of the second competition parameter and the actual sample click-through rate used in the i-th round, the first loss value in the i-th round is determined, wherein the actual sample click-through rate used in the i-th round is used to indicate whether the sample account used in the i-th round actually clicks on the sample media resource used in the i-th round after the sample media resource used in the i-th round is published to the sample account used in the i-th round; Based on the sample account feature vector determined in the i-th round, the sample media resource feature vector determined in the i-th round, and the number of sample virtual resources used in the i-th round, the value of the first competition parameter in the i-th round is determined, and based on the value of the second competition parameter in the i-th round and the value of the first competition parameter in the i-th round, the second loss value in the i-th round is determined. The target loss value for the i-th round is determined based on the first loss value and the second loss value for the i-th round. When the target loss value in the i-th round does not meet the target loss condition, the parameters in at least one of the initial account tower model structure and the initial media resource tower model structure obtained after the (i-1)-th round of training are adjusted.

8. A media resource publishing device, characterized in that, include: The acquisition module is used to acquire the target account feature vector, the target media resource feature vector, and the target virtual resource quantity of the target account. The target media resource feature vector includes a set of media resource features of the target media resource, and the target virtual resource quantity is the number of virtual resources that need to be transferred out to publish the target media resource to the target account. The processing module is used to logically process the target account feature vector, the target media resource feature vector, and the target virtual resource quantity through the target prediction network structure in the target prediction neural network model to obtain the target value of the first competition parameter of the target media resource. The target value of the first competition parameter of the target media resource is used to determine whether to select the target media resource in the target media resource set as a media resource to be published to the target account. The target loss value corresponding to the target prediction neural network model satisfies a preset target loss condition. The target loss value is a loss value determined based on a first loss value and a second loss value. The first loss value is a loss value determined based on the sample account feature vector and the sample media resource feature vector. The second loss value is a loss value determined based on the first sample prediction result and the second sample prediction result. The first sample prediction result is the value of the second competition parameter determined based on the sample account feature vector and the sample media resource feature vector. The second sample prediction result is the value of the first competition parameter determined based on the sample account feature vector, the sample media resource feature vector, and the corresponding sample virtual resource quantity. The publishing module is used to publish the target media resource to the target account when the target value of the first competitive parameter of the target media resource meets the preset target recall condition, wherein the target recall condition is used to determine the media resource to be published to the target account in the target media resource set.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein the program, when executed, performs the method described in any one of claims 1 to 7.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 7.

11. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to execute the method described in any one of claims 1 to 7 through the computer program.

Images