Media resource conversion attribution prediction model generation, application method and device

Abstract

This disclosure relates to a method and apparatus for generating and applying a media resource conversion attribution prediction model. The method includes: acquiring a media resource conversion attribution prediction model; the model takes media resource profile description information and user profile description information as input, and outputs a first probability and a second probability, where the first probability represents the probability that the attribution result is user-triggered conversion, and the second probability represents the probability that the attribution result is media resource exposure conversion; acquiring a training sample set, which includes a first positive sample, a second positive sample, and a negative sample, each of which includes a first label; inputting the training sample set into the model to obtain the first probability and the second probability; determining a loss value based on the first probability, the second probability, and the first label; and adjusting the parameters of the media resource conversion attribution prediction model based on the loss value until a training stopping condition is met. This disclosure can improve attribution accuracy.

Description

Technical Field

[0001] This disclosure relates to the field of computer applications, and in particular to the generation, application method and apparatus of media resource conversion attribution prediction model. Background Technology

[0002] Media resources are ubiquitous in life. Many media platforms can recommend and display various media resources to users. Using a media resource delivery system, various media resources can be delivered to users according to the requirements of the media resource owner.

[0003] Media resource conversion largely reflects the value created by media resources. Generally, actions such as orders, downloads, activations, or recharges resulting from media resource placement are considered conversions. After placing media resources, the media resource owner needs to analyze the reasons for these conversions; this is called media resource attribution analysis. Attribution analysis is a crucial step in evaluating the effectiveness of media resource placement, and its results are also an important basis for media resource recommendations. However, the effectiveness of media resource attribution in related technologies needs improvement; the attribution results are not yet accurate enough, thus affecting the effectiveness of media resource placement. Summary of the Invention

[0004] This disclosure provides a method and apparatus for generating and applying a media resource conversion attribution prediction model, in order to at least solve the problem of inaccurate media resource attribution results in related technologies. The technical solution of this disclosure is as follows:

[0005] According to a first aspect of the present disclosure, a method for generating a media resource conversion attribution prediction model is provided, comprising:

[0006] A media resource conversion attribution prediction model is obtained; the media resource conversion attribution prediction model takes media resource profile description information and user profile description information as input, and takes a first probability and a second probability as output. The first probability represents the probability that the attribution result is triggered by the user and the second probability represents the probability that the attribution result is converted by the media resource exposure.

[0007] Obtain a training sample set, which includes a first positive sample, a second positive sample, and a negative sample. The first positive sample, the second positive sample, and the negative sample all include media resource profile description information, user profile description information, and a first tag. The first tag of the first positive sample represents that the attribution result is user-triggered conversion. The first tag of the second positive sample represents that the attribution result is media resource exposure conversion. The first tag of the negative sample represents that the media resource has not been converted.

[0008] The training samples are input into the media resource conversion attribution prediction model so that the media resource conversion attribution prediction model obtains a first probability and a second probability based on the media resource profile description information and user profile description information of the training samples.

[0009] The loss value is determined based on the first probability, the second probability, and the first label of the training sample;

[0010] The parameters of the media resource conversion attribution prediction model are adjusted based on the loss value until the training stopping condition is met.

[0011] In one exemplary embodiment, the media resource conversion attribution prediction model includes an information receiving layer, an intermediate layer, and an output layer;

[0012] The intermediate layer is a flat structure consisting of a first intermediate layer, a second intermediate layer, and a third intermediate layer; the first intermediate layer, the second intermediate layer, and the third intermediate layer respectively output the trigger conversion rate of media resources, the user's trigger rate of media resources, and the non-trigger conversion rate of media resources.

[0013] The output layer is a flat structure composed of a first output layer and a second output layer; the first output layer and the second output layer respectively output a first probability and a second probability.

[0014] The first intermediate layer, the second intermediate layer, and the third intermediate layer are all connected to the information receiving layer; the first intermediate layer and the second intermediate layer are both connected to the first output layer; and the second intermediate layer and the third intermediate layer are both connected to the second output layer.

[0015] In one exemplary embodiment, the information receiving layer includes a feature data embedding layer and a feature output layer connected together, and the first intermediate layer, the second intermediate layer, and the third intermediate layer are all connected to the feature output layer;

[0016] The feature data embedding layer is used to embed sparse feature data in the media resource profile description information and the user profile description information to obtain the first embedded feature; the sparse feature data includes media resource dimension features, user dimension features and user-media resource cross features.

[0017] The feature output layer is used to receive the first embedded feature and the second embedded feature, wherein the second embedded feature is dense feature data in the media resource profile description information and the user profile description information.

[0018] In an exemplary implementation, obtaining the first probability and the second probability based on the media resource profile description information and user profile description information of the training samples includes:

[0019] The first intermediate layer outputs the trigger conversion rate of the media resource based on the media resource profile description information and the user profile description information;

[0020] The second intermediate layer outputs the user's trigger rate for media resources based on the media resource profile description information and the user profile description information;

[0021] The third intermediate layer outputs the non-triggered conversion rate of media resources based on the media resource profile description information and the user profile description information;

[0022] The first output layer calculates the first product of the trigger conversion rate of the media resource output by the first intermediate layer and the user trigger rate of the media resource output by the second intermediate layer; the first product is determined as the first probability.

[0023] The second output layer obtains an intermediate result based on the user trigger rate of media resources output by the second intermediate layer, calculates the second product of the intermediate result and the media resource non-trigger conversion rate output by the third intermediate layer, and determines the second product as the second probability.

[0024] In one exemplary implementation, determining a loss value based on the first probability, the second probability, and the first label of the training sample includes:

[0025] Calculate the first loss based on the first probability and the first label of the training sample;

[0026] The second loss is calculated based on the second probability and the first label of the training sample;

[0027] The loss value is calculated based on the first loss and the second loss.

[0028] In an exemplary implementation, the training samples in the training sample set further include a second label, which is used to characterize whether a user-triggered action exists. Accordingly, determining the loss value based on the first probability, the second probability, and the first label of the training samples includes:

[0029] Calculate the first loss based on the first probability and the first label of the training sample;

[0030] The second loss is calculated based on the second probability and the first label of the training sample;

[0031] The third loss is calculated based on the user's trigger rate for media resources and the second label of the training samples;

[0032] The loss value is calculated based on the first loss, the second loss, and the third loss.

[0033] In one exemplary implementation, the loss value is obtained by weighted summation of the individual losses it comprises.

[0034] According to a second aspect of the present disclosure, a method for applying a media resource conversion attribution prediction model is provided, comprising:

[0035] Obtain media resource profile description information and user profile description information;

[0036] Input the media resource profile description information and user profile description information into the media resource conversion attribution prediction model to obtain the first probability, the second probability and / or the media resource exposure conversion rate.

[0037] The media resource conversion attribution prediction model is obtained based on the media resource conversion attribution prediction model generation method described in any of the above embodiments, and the media resource exposure conversion rate is the sum of the first probability and the second probability.

[0038] According to a third aspect of the present disclosure, a media resource push method is provided, characterized in that it includes:

[0039] Obtain media resource requests and extract user profile description information from the media resource requests;

[0040] Obtain a media resource set and extract the media resource profile description information for each media resource in the media resource set;

[0041] For each media resource in the media resource set, the media resource profile description information and the user profile description information are input into the media resource conversion attribution prediction model to obtain the media resource exposure conversion rate of the media resource relative to the user; the media resource exposure conversion rate is the sum of the first probability and the second probability output by the media resource conversion attribution prediction model.

[0042] The target media resources are determined based on the media resource exposure conversion rate of each media resource in the media resource set.

[0043] Push the target media resources;

[0044] The media resource conversion attribution prediction model is obtained based on the media resource conversion attribution prediction model generation method described in any of the above embodiments.

[0045] According to a fourth aspect of the present disclosure, a media resource conversion attribution prediction model generation apparatus is provided, comprising:

[0046] The media resource conversion attribution prediction model acquisition module is configured to acquire the media resource conversion attribution prediction model. The media resource conversion attribution prediction model takes media resource profile description information and user profile description information as inputs and takes a first probability and a second probability as outputs. The first probability represents the probability that the attribution result is triggered by the user and the second probability represents the probability that the attribution result is converted by the media resource exposure.

[0047] The training sample set acquisition module is configured to acquire a training sample set, which includes a first positive sample, a second positive sample, and a negative sample. Each of the first positive sample, the second positive sample, and the negative sample includes media resource profile description information, user profile description information, and a first tag. The first tag of the first positive sample represents that the attribution result is user-triggered conversion. The first tag of the second positive sample represents that the attribution result is media resource exposure conversion. The first tag of the negative sample represents that the media resource has not been converted.

[0048] The training prediction module is configured to input the training samples into the media resource conversion attribution prediction model, so that the media resource conversion attribution prediction model obtains a first probability and a second probability based on the media resource profile description information and user profile description information of the training samples.

[0049] The loss value determination module is configured to determine a loss value based on the first probability, the second probability, and the first label of the training sample;

[0050] The training control module is configured to adjust the parameters of the media resource conversion attribution prediction model based on the loss value until the training stop condition is met.

[0051] In one exemplary embodiment, the media resource conversion attribution prediction model includes an information receiving layer, an intermediate layer, and an output layer;

[0052] The intermediate layer is a flat structure consisting of a first intermediate layer, a second intermediate layer, and a third intermediate layer; the first intermediate layer, the second intermediate layer, and the third intermediate layer respectively output the trigger conversion rate of media resources, the user's trigger rate of media resources, and the non-trigger conversion rate of media resources.

[0053] The output layer is a flat structure composed of a first output layer and a second output layer; the first output layer and the second output layer respectively output a first probability and a second probability.

[0054] The first intermediate layer, the second intermediate layer, and the third intermediate layer are all connected to the information receiving layer; the first intermediate layer and the second intermediate layer are both connected to the first output layer; and the second intermediate layer and the third intermediate layer are both connected to the second output layer.

[0055] In one exemplary embodiment, the information receiving layer includes a feature data embedding layer and a feature output layer connected together, and the first intermediate layer, the second intermediate layer, and the third intermediate layer are all connected to the feature output layer;

[0056] The feature data embedding layer is used to embed sparse feature data in the media resource profile description information and the user profile description information to obtain the first embedded feature; the sparse feature data includes media resource dimension features, user dimension features and user-media resource cross features.

[0057] The feature output layer is used to receive the first embedded feature and the second embedded feature, wherein the second embedded feature is dense feature data in the media resource profile description information and the user profile description information.

[0058] In one exemplary implementation, the training prediction module is configured to perform:

[0059] The first intermediate layer outputs the trigger conversion rate of the media resource based on the media resource profile description information and the user profile description information;

[0060] The second intermediate layer outputs the user's trigger rate for media resources based on the media resource profile description information and the user profile description information;

[0061] The third intermediate layer outputs the non-triggered conversion rate of media resources based on the media resource profile description information and the user profile description information;

[0062] The first output layer calculates the first product of the trigger conversion rate of the media resource output by the first intermediate layer and the user trigger rate of the media resource output by the second intermediate layer; the first product is determined as the first probability.

[0063] The second output layer obtains an intermediate result based on the user trigger rate of media resources output by the second intermediate layer, calculates the second product of the intermediate result and the non-trigger conversion rate of media resources output by the third intermediate layer, and determines the second product as the second probability.

[0064] In one exemplary embodiment, the loss value determination module includes:

[0065] The first loss calculation unit is configured to perform a first loss calculation based on the first probability and the first label of the training sample;

[0066] The second loss calculation unit is configured to perform a second loss calculation based on the second probability and the first label of the training sample;

[0067] The first loss totaling unit is configured to perform the calculation of the loss value based on the first loss and the second loss.

[0068] In one exemplary embodiment, the loss value determination module includes:

[0069] The first loss calculation unit is configured to perform calculation of a first loss based on the first probability and the first label of the training sample;

[0070] The second loss calculation unit is configured to perform a second loss calculation based on the second probability and the first label of the training sample;

[0071] The third loss calculation unit is configured to perform a third loss calculation based on the user's trigger rate on the media resource and the second label of the training sample;

[0072] The second loss totaling unit calculates the loss value based on the first loss, the second loss, and the third loss.

[0073] In one exemplary implementation, the loss value is obtained by weighted summation of the individual losses it comprises.

[0074] According to a fifth aspect of the present disclosure, a media resource conversion attribution prediction model application apparatus is provided, comprising:

[0075] The information acquisition module is configured to acquire media resource profile description information and user profile description information;

[0076] The processing module is configured to input the media resource profile description information and user profile description information into the media resource conversion attribution prediction model to obtain a first probability, a second probability, and / or the media resource exposure conversion rate.

[0077] The media resource conversion attribution prediction model is obtained using the media resource conversion attribution prediction model generation method described in any of the above embodiments, and the media resource exposure conversion rate is the sum of the first probability and the second probability.

[0078] According to a sixth aspect of the present disclosure, a media resource push device is provided, comprising:

[0079] The media resource request acquisition module is configured to acquire media resource requests and extract user profile description information from the media resource requests;

[0080] The media resource set determination module is configured to acquire the media resource set and extract the media resource profile description information of each media resource in the media resource set;

[0081] The media resource exposure conversion rate determination module is configured to perform, for each media resource in the media resource set, input the media resource profile description information and the user profile description information into the media resource conversion attribution prediction model to obtain the media resource exposure conversion rate of the media resource relative to the user; the media resource exposure conversion rate is the sum of the first probability and the second probability output by the media resource conversion attribution prediction model.

[0082] The target media resource determination module is configured to determine the target media resources based on the media resource exposure conversion rate corresponding to each media resource in the media resource set;

[0083] The push module is configured to push the target media resource;

[0084] The media resource conversion attribution prediction model is obtained using the media resource conversion attribution prediction model generation method described in any of the above embodiments.

[0085] According to a seventh aspect of the present disclosure, an electronic device is provided, comprising:

[0086] processor;

[0087] Memory used to store the processor's executable instructions;

[0088] The processor is configured to execute the instructions to implement the media resource conversion attribution prediction model generation method, media resource conversion attribution prediction model application method, or media resource push method as described in any of the above embodiments.

[0089] According to an eighth aspect of the present disclosure, a storage medium is provided that, when instructions in the storage medium are executed by a processor of an electronic device, enables the electronic device to perform the media resource conversion attribution prediction model generation method, the media resource conversion attribution prediction model application method, or the media resource push method described in any of the above embodiments.

[0090] According to a ninth aspect of the present disclosure, a computer program product is provided, the computer program product including computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium, and executes the computer instructions, causing the electronic device to perform the media resource conversion attribution prediction model generation method, media resource conversion attribution prediction model application method, or media resource push method provided in any of the above embodiments.

[0091] The technical solutions provided by the embodiments of this disclosure have at least the following beneficial effects:

[0092] By designing the structure of the media resource conversion attribution prediction model, each intermediate layer can independently output a factor used to obtain the attribution result. The first output layer, used to calculate the first probability, is only connected to the intermediate layers that output factors related to the first probability, and the second output layer, used to calculate the second probability, is only connected to the intermediate layers that output factors related to the second probability. This ensures that the configuration of each logical layer and the connections between them are largely compatible with the calculation methods for the first and second probabilities. This rational design reduces model complexity and improves robustness. The construction and training process of the media resource conversion attribution prediction model considers both user exposure conversion and user-triggered conversion as attribution results. Compared to related technologies that simply attribute media resources to either user-triggered conversion or media resource exposure conversion, resulting in low attribution accuracy, this disclosure can simultaneously output both probabilities of user-triggered conversion and media resource exposure conversion, leading to more accurate attribution and significantly improved attribution accuracy.

[0093] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0094] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0095] Figure 1 This is an application environment diagram illustrating a media resource conversion attribution prediction model generation method according to an exemplary embodiment;

[0096] Figure 2 This is a flowchart illustrating a method for generating a media resource conversion attribution prediction model according to an exemplary embodiment;

[0097] Figure 3 This is a schematic diagram of a feasible structure for a media resource conversion attribution prediction model, according to an exemplary embodiment.

[0098] Figure 4 This is a flowchart of step S30 in a media resource conversion attribution prediction model generation method according to an exemplary embodiment;

[0099] Figure 5 This is a flowchart of step S40 in a media resource conversion attribution prediction model generation method according to an exemplary embodiment;

[0100] Figure 6 This is a flowchart of step S40 in a media resource conversion attribution prediction model generation method according to an exemplary embodiment;

[0101] Figure 7 This is a flowchart illustrating an application method of a media resource conversion attribution prediction model according to an exemplary embodiment;

[0102] Figure 8 This is a flowchart illustrating a media resource delivery method according to an exemplary embodiment;

[0103] Figure 9 This is a block diagram illustrating a media resource conversion attribution prediction model generation apparatus according to an exemplary embodiment;

[0104] Figure 10 This is a block diagram illustrating an application device for a media resource conversion attribution prediction model according to an exemplary embodiment;

[0105] Figure 11 This is a block diagram illustrating a media resource delivery device according to an exemplary embodiment;

[0106] Figure 12 This is a block diagram illustrating an electronic device that performs a method for generating a media resource conversion attribution prediction model, a method for applying a media resource conversion attribution prediction model, or a method for pushing media resources, according to an exemplary embodiment. Detailed Implementation

[0107] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0108] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure 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 this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0109] The following explanations of some terms used in the embodiments of this disclosure are provided to facilitate understanding by those skilled in the art.

[0110] A feed stream is a content aggregator that combines several message sources actively subscribed to by the user, helping them continuously access the latest content from their subscriptions. Each object (such as a status or message) in a feed stream constitutes a feed. Feed streams can be displayed in various ways, such as in chronological order of updates or by weighting based on certain factors. For example, in short video applications, the feed stream might contain short videos posted by accounts the user follows or media resources pushed to the user. Users can continuously browse the feed stream by tapping and swiping.

[0111] User triggers: User actions, such as clicking or swiping, are considered triggers. Clicks can be single or double clicks, and swipes can be upward, downward, leftward, or rightward swipes, etc. There are no restrictions here. For example, if a user clicks on a media resource and generates an order, download, activation, or recharge for the media resource owner, then the above user triggers result in media resource conversion.

[0112] Media resource conversion: Generally, the orders, downloads, activations, or recharges resulting from the placement of media resources are referred to as conversions. For example, media resource conversion can be one or more of the following: purchase operations, download operations, page redirects, and recharge operations triggered by media resource exposure.

[0113] Media resource exposure: Media resources are delivered to users. In this disclosure, media resource exposure is a prerequisite for media resource conversion.

[0114] OCPC: Optimized Cost Per Click, is a media resource operation model that optimizes click bidding based on target conversion. It helps media resource owners acquire more high-quality traffic while improving conversion completion rates.

[0115] Secondary retention: Among the new users added on the same day, those who log in again on the first day after the initial login.

[0116] ROI: Return on Investment; refers to the value that should be returned through an investment.

[0117] Deep payment: Using user payment as the criterion for media resource conversion, that is, media resource conversion is only determined when users pay.

[0118] AUC: Area Under Curve; defined as the area under the receiver operating characteristic curve. AUC is a performance metric for evaluating the quality of a learner.

[0119] In related technologies, there are various methods for media resource delivery, which is one reason for the complexity of media resource attribution. Taking feed-based media resource delivery as an example, feed-based media resource delivery includes dual-cascade streaming media resource delivery and direct media resource delivery. In the dual-cascade streaming media resource delivery method, the first hop is triggered by the user to play the media resource video, and the second hop is triggered again by the user to display the media resource details page or media resource download page. In the direct media resource delivery method, the media resource video is played directly, and the media resource details page or media resource download page is displayed after a single user trigger. Dual-cascade streaming media resource delivery can use the first hop as the attribution result for media resource conversion, while direct media resource delivery can use media resource exposure as the attribution result for media resource conversion.

[0120] In related technologies, to obtain intuitive attribution results and simplify attribution complexity, attribution logic is often placed at the forefront. For example, the most recent media resource exposure leading to a media resource conversion can be used as the attribution result for modeling. This might lead to a media resource conversion attribution prediction model that prioritizes media resource exposure conversion while downplaying user-triggered conversion. However, in many scenarios, the number of media resource exposures far exceeds the number of media resource conversions, resulting in sparse training data for such models and significantly impacting their accuracy. Alternatively, a single user trigger leading to a media resource conversion can be used as the attribution result for modeling. This might also lead to a media resource conversion attribution prediction model that prioritizes user-triggered conversion while downplaying media resource exposure conversion. In relevant statistics, media resource exposure may be used as the attribution result in over 40% of cases; ignoring this situation clearly has a significant impact on the accuracy of media resource conversion attribution prediction models. In summary, the media resource attribution modeling schemes in related technologies are unable to comprehensively consider the attribution of multiple media resources, and can only focus on the attribution of a single media resource. As a result, the accuracy of the media resource attribution prediction model is low and the attribution results are poor.

[0121] However, as the requirements for the deep backend performance of media resources become increasingly demanding, for example, the requirements for performance indicators such as OCPC, second-day retention rate, and ROI are becoming more stringent, and the product forms of media resource delivery are becoming more diversified, there is an urgent need for a highly accurate media resource conversion attribution prediction model to provide highly accurate attribution results under various circumstances.

[0122] This disclosure provides a media resource conversion attribution prediction model generation scheme. The media resource conversion attribution prediction model obtained by the above generation scheme takes media resource profile description information and user profile description information as input. It can not only output the probability of media resource exposure conversion as an attribution result, but also the probability of user triggering conversion as an attribution result. Furthermore, by designing the structure and training process of the media resource conversion attribution prediction model, attribution accuracy can be guaranteed.

[0123] Please see Figure 1 The diagram illustrates an application environment for a media resource conversion attribution prediction model generation method according to an exemplary embodiment. The application environment may include a terminal 110, a media resource server 120, and a modeling server 130. Figure 1 The media resource server 120 and model server 130 shown can be independent physical servers, or they can be a server cluster or a distributed system composed of multiple physical servers. The media resource server 120 and the model server 130 can be the same server or different servers. The terminal 110 and the media resource server 120, as well as the model server 130 and the media resource server 120, can be connected via wired or wireless networks.

[0124] The terminal 110 can be a smartphone, tablet, laptop, desktop computer, etc., but is not limited to these. The terminal 110 can run a client that receives background services from the media resource server 120. This client can obtain user profile description information or media resource-related behavioral information for the user corresponding to the terminal 110. For example, if a user plays a media resource, media resource exposure information is generated; if a user clicks on the media resource, media resource trigger information is generated; if a user enters the media resource's details page or download page, media resource conversion information is generated. The client then uploads the aforementioned media resource-related behavioral information to the media resource server 120.

[0125] The aforementioned media resource server 120 is used to provide media resource services to users and manage user profile description information and media resource profile description information. The media resource profile description information may include user behavioral information related to media resources. For example, the media resource server 120 can obtain media resource dimensional features based on the user's behavioral information related to media resources, and may also obtain user-media resource cross-features. The aforementioned user profile description information may include identifiers, gender, age, preferences, etc. The media resource server 120 can obtain user dimensional features based on the aforementioned user profile description information, and may also obtain user-media resource cross-features.

[0126] Modeling server 130 is used to construct a media resource conversion attribution prediction model. It obtains user profile description information and media resource profile description information through communication with media resource server 120. Based on the user profile description information and media resource profile description information, it trains and updates the media resource conversion attribution prediction model and provides media resource conversion attribution services to media resource server 120. Correspondingly, relying on the media resource conversion attribution services provided by modeling server 130, media resource server 120 can provide users with media resource recommendation services or media resource data processing services.

[0127] The media resource conversion attribution prediction model constructed in the aforementioned modeling server 130 is a multi-task, multi-label model that can output multiple results, each of which corresponds to the probability of an attribution result, thereby enabling more accurate media resource attribution.

[0128] Figure 2 This is a flowchart illustrating a method for generating a media resource conversion attribution prediction model according to an exemplary embodiment, such as... Figure 2 As shown, the media resource conversion attribution prediction model generation method is applied to... Figure 1 The following steps are described in the modeling server 130 shown.

[0129] In step S10, a media resource conversion attribution prediction model is obtained. The media resource conversion attribution prediction model takes media resource profile description information and user profile description information as inputs and a first probability and a second probability as outputs. The first probability represents the probability that the attribution result is triggered by the user, and the second probability represents the probability that the attribution result is converted by the media resource exposure.

[0130] In some feasible implementations, the aforementioned media resources can be delivered based on a feed stream, or by embedding in a webpage, or by being delivered alongside multimedia resources. This disclosure does not limit the delivery method of the aforementioned media resources.

[0131] In step S20, a training sample set is obtained, which includes a first positive sample, a second positive sample, and a negative sample. Each of the first positive sample, the second positive sample, and the negative sample includes media resource profile description information, user profile description information, and a first tag. The first tag of the first positive sample represents the attribution result as user-triggered conversion, the first tag of the second positive sample represents the attribution result as media resource exposure conversion, and the first tag of the negative sample represents that the media resource has not been converted.

[0132] In some feasible implementations, the training sample set may specifically include a first positive sample, a second positive sample, and a negative sample. The first positive sample, the second positive sample, and the negative sample all include media resource profile description information, user profile description information, a first tag, and a second tag. The first tag of the first positive sample represents the attribution result as user-triggered conversion, the first tag of the second positive sample represents the attribution result as media resource exposure conversion, and the first tag of the negative sample represents that the media resource was not converted. The second tags of the first positive sample, the second positive sample, and the negative sample all represent whether a user-triggered action exists.

[0133] For example, the training sample set mentioned above can be derived from the media resource server 120.

[0134] In some feasible implementations, each sample in the training sample set may include media resource profile description information, user profile description information, a first label, and a second label. The value of the first label represents the attribution result of the sample, and the second label represents whether the user corresponding to the user profile description information has triggered behavior for the media resource corresponding to the media resource profile description information.

[0135] For example, for the first positive sample, its first label represents the attribution result as user-triggered conversion, and its scvr1=1, scvr0=0; for the second positive sample, its first label represents the attribution result as media resource exposure conversion, and its scvr1=0, scvr0=1; for the negative sample, its first label represents that the media resource was not converted, and its scvr1=0, scvr0=0.

[0136] For example, for the first positive sample, the second positive sample, and the negative sample, their second label all represent an action attribute. This action attribute indicates whether the media resource in the training sample was triggered by the user in that sample (regardless of whether it was converted, only the existence of user triggering is considered). For example, if the media resource in the sample was triggered by the user in that sample, then the second label is 1; otherwise, it is 0.

[0137] In some feasible implementations, in order to improve the training effect of the media resource conversion attribution prediction model, the ratio of the number of first positive samples, the number of second positive samples, and the number of negative samples can be further set to ensure that the training process of the media resource conversion attribution prediction model fully takes into account the attribution situations corresponding to the two positive samples and has good attribution ability for both attribution situations.

[0138] In step S30, the training samples are input into the media resource conversion attribution prediction model so that the media resource conversion attribution prediction model obtains a first probability and a second probability based on the media resource profile description information and user profile description information of the training samples.

[0139] In some feasible implementations, the triggered conversion rate of media resources, the user's trigger rate for media resources, and the non-triggered conversion rate of media resources can be obtained from the media resource conversion attribution prediction model; and a first probability and a second probability can be obtained based on the aforementioned triggered conversion rate of media resources, user's trigger rate for media resources, and non-triggered conversion rate of media resources.

[0140] In some feasible implementations, the media resource conversion attribution prediction model can obtain three types of data based on the input media resource profile description information and user profile description information: the trigger conversion rate of media resources, the user's trigger rate for media resources, and the non-trigger conversion rate of media resources. Based on these three types of data, the first probability and the second probability are calculated respectively.

[0141] In the embodiments disclosed herein, pcvr1 is used to represent the trigger conversion rate of media resources, pctr is used to represent the user's trigger rate of media resources, and pcvr0 is used to represent the non-trigger conversion rate of media resources.

[0142] In step S40, the loss value is determined based on the first probability, the second probability, and the first label of the training sample.

[0143] In some feasible implementations, the loss value can be determined specifically based on the first probability, the second probability, the user's trigger rate for media resources, and the first and second labels of the training samples.

[0144] In step S50, the parameters of the media resource conversion attribution prediction model are adjusted based on the aforementioned loss value until the training stopping condition is met.

[0145] In some feasible implementations, the media resource conversion attribution prediction model is trained using training samples in steps S30-S50. This disclosure can use various feasible training methods such as gradient descent, stochastic gradient descent, and batch gradient descent. The embodiments of this disclosure do not limit the specific training method.

[0146] This disclosure does not limit the training stopping condition. In one feasible implementation, the training stopping condition can be determined based on a loss threshold. For example, if the loss value is greater than or equal to the loss threshold, the parameters of the media resource conversion attribution prediction model are adjusted until the loss value generated by the media resource conversion attribution prediction model is less than the loss threshold. When the loss value generated by the media resource conversion attribution prediction model is less than the loss threshold, it indicates that the media resource conversion attribution prediction model has met the requirements and can be applied. The loss threshold can be a value set according to requirements, such as 0.1, but it is not a specific limitation of this disclosure.

[0147] In another feasible implementation, the training stopping condition can be determined based on the number of iterations. Each time the step is performed: the parameters of the media resource conversion attribution prediction model are adjusted according to the aforementioned loss value, the number of iterations increments by one, until the number of iterations reaches a preset iteration threshold. This iteration threshold can be a value set according to requirements, but is not intended as a specific limitation of this disclosure.

[0148] In another feasible implementation, the number of iterations and the loss threshold can also be used together to determine the training stopping condition, which will not be elaborated in the embodiments of this disclosure.

[0149] The media resource conversion attribution prediction model generation method shown in this disclosure can output the attribution result as the probability of user-triggered conversion or the probability of media resource exposure conversion. Compared with the shortcomings of related technologies that only consider user-triggered conversion or media resource exposure conversion as attribution results, resulting in low attribution accuracy, this disclosure considers both attribution results in a compatible manner, obtaining the probability of a media resource conversion being attributed to media resource exposure conversion and the probability of being attributed to user-triggered conversion, respectively, thereby accurately obtaining the probability of each attribution result corresponding to the media resource conversion.

[0150] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0151] In some feasible implementations, such as Figure 3 As shown, it illustrates a feasible structural diagram of a media resource conversion attribution prediction model according to an exemplary embodiment.

[0152] The aforementioned media resource conversion attribution prediction model includes an information receiving layer 01, an intermediate layer 02, and an output layer 03. The information receiving layer 01 receives media resource profile description information and user profile description information, so that the output layer 03 of the aforementioned media resource conversion attribution prediction model can ultimately output a first probability scvr1 and a second probability scvr0 based on the aforementioned media resource profile description information and the aforementioned user profile description information.

[0153] The aforementioned intermediate layer 02 is a flat structure composed of a first intermediate layer 021, a second intermediate layer 022, and a third intermediate layer 023. The first intermediate layer 021, the second intermediate layer 022, and the third intermediate layer 023 respectively correspond to the trigger conversion rate pcvr1 of the output media resources, the user trigger rate pctr of the media resources, and the non-trigger conversion rate pcvr0 of the media resources. The first intermediate layer 021, the second intermediate layer 022, and the third intermediate layer 023 are all connected to the aforementioned information receiving layer 01.

[0154] This disclosure does not limit the specific structures of the first intermediate layer 021, the second intermediate layer 022, and the third intermediate layer 023. The specific structures of the first intermediate layer 021, the second intermediate layer 022, and the third intermediate layer 023 can be the same or different. In some feasible implementations, any one of the first intermediate layer 021, the second intermediate layer 022, and the third intermediate layer 023 can be configured as a fully connected layer or a cascaded fully connected layer, and the number of neurons in any fully connected layer constituting the first, second, and third intermediate layers can also be configured according to actual needs. This disclosure improves the flexibility of configuring the first, second, and third intermediate layers by supporting free settings for fully connected layers, facilitating a structurally compatible media resource conversion attribution prediction model with its actual application scenarios.

[0155] For example, please refer to Figure 3 The first intermediate layer 021, the second intermediate layer 022, and the third intermediate layer 023 all have the same structure. Taking the first intermediate layer 021 as an example, it includes two cascaded fully connected layers. The first fully connected layer 021, which is closer to the information receiving layer 01, includes 512 neurons, and the second fully connected layer 022, which is farther from the information receiving layer 01, includes 256 neurons. Correspondingly, the second intermediate layer 022 includes the third fully connected layer 0221 and the fourth fully connected layer 0222, and the third intermediate layer 023 includes the fifth fully connected layer 0231 and the sixth fully connected layer 0232.

[0156] The aforementioned output layer 03 is a flat structure composed of a first output layer 031 and a second output layer 032; the first output layer 031 and the second output layer 032 respectively output the first probability and the second probability. The first intermediate layer 021 and the second intermediate layer 022 are both connected to the first output layer 031; the second intermediate layer 022 and the third intermediate layer 023 are both connected to the second output layer 032.

[0157] In a feasible implementation, a first multiplier 0311 can be set in the first output layer 031. The first multiplier 0311 is used to calculate the first product of the output of the first intermediate layer 021 and the output of the second intermediate layer 022. The first product represents the first probability scvr1, that is, the first probability scvr1 = trigger conversion rate pcvr1 * user trigger rate pctr for media resources.

[0158] In the second output layer 032, a subtractor 0321 and a multiplier 0322 are configured. The subtractor 0321 is used to obtain an intermediate result based on the output of the second intermediate layer 022. The second multiplier 0322 is used to calculate the second product of the intermediate result and the output of the third intermediate layer 023. The second product represents the second probability scvr0. Specifically, the subtractor is used to calculate the probability that the user will not trigger the media resource. Therefore, the subtractor performs the operation of (1 - the output of the second intermediate layer). That is, the second probability scvr0 = (1 - the user's trigger rate for the media resource pctr) * the media resource non-trigger conversion rate pcvr0.

[0159] For example, please refer to Figure 3 The first probability scvr1 is obtained by passing the pcvr1 output from the first intermediate layer 021 and the pctr output from the second intermediate layer 022 through the first multiplier 0311 set in the first output layer 031; the probability that the user will not trigger the media resource is obtained by passing the pctr output from the second intermediate layer 023 through the subtractor 0321 set in the second output layer 032, i.e. (1-pctr). The probability that the user will not trigger the media resource is obtained by passing the probability that the user will not trigger the media resource output from the subtractor 0322 and the pcvr0 output from the third intermediate layer 032 through the second multiplier 0322 set in the second output layer 032.

[0160] In some feasible embodiments, the information receiving layer 01 includes a feature data embedding layer 011 and a feature output layer 012 connected to each other, and the first intermediate layer 021, the second intermediate layer 022 and the third intermediate layer 023 are all connected to the feature output layer 012.

[0161] For example, the aforementioned feature data embedding layer 011 is used to embed sparse feature data in the aforementioned media resource profile description information and user profile description information to obtain a first embedded feature. The feature data embedding layer 011 can obtain a dense first embedded feature by reducing the dimensionality of the sparse feature data. The function of the feature data embedding layer 011 is to reduce the dimensionality of the input data of the media resource conversion attribution prediction model, thereby improving the expressive power of the media resource conversion attribution prediction model for the input data and saving storage and computing resources. This embodiment does not limit the specific structure of the feature data embedding layer 011.

[0162] In some feasible implementations, the aforementioned sparse feature data includes media resource dimension features, user dimension features, and user-media resource cross features.

[0163] For example, the above-mentioned media resource dimension features can describe the relevant information of media resources in the media resource dimension. This disclosure does not limit the specific content of the media resource dimension features. For example, the above-mentioned media resource dimension features may include, but are not limited to, one or more of the following: media resource identifier, media resource main identifier, media resource main name, media resource main industry category, media resource size, media resource summary, and media resource type.

[0164] For example, the above-mentioned user dimension features can describe relevant information about users in the user dimension. This disclosure does not limit the specific content of user dimension features. For example, the above-mentioned user dimension features may include, but are not limited to, one or more of the following: user identifier, age, gender, region, IP address, device model, media resources clicked in the past, and media resources activated in the past.

[0165] For example, the aforementioned user media resource cross-features can describe user behavior information or preference information relative to media resources. This disclosure does not limit the specific content of the user media resource cross-features. For example, the aforementioned user media resource cross-features may include, but are not limited to, one or more of the user's historical click behavior, historical activation behavior, and preference marking behavior on that media resource.

[0166] In some feasible implementations, the feature output layer 012 can be directly connected to the feature data embedding layer 011 to receive the first embedded feature and output the first embedded feature to the first intermediate layer 021, the second intermediate layer 022 and the third intermediate layer 023.

[0167] In other feasible implementations, to enhance the richness of the input data for the media resource conversion attribution prediction model and thus improve its accuracy, the feature output layer 012 can receive a first embedded feature by connecting to the feature data embedding layer, or it can directly receive a second embedded feature. The combination of the first and second embedded features is then output to the first intermediate layer 021, the second intermediate layer 022, and the third intermediate layer 023. The second embedded feature is dense feature data from the media resource profile description information and the user profile description information.

[0168] In this disclosure, the second embedded feature is dense feature data that does not require dimensionality reduction through the feature data embedding layer 011. This disclosure does not limit the specific content of the second embedded feature. For example, the second embedded feature may include, but is not limited to, features representing media resources that users do not like, embedded features obtained based on users browsing natural works, and embedded features of media resources obtained based on visual algorithms, or one or more of these.

[0169] Accordingly, in some feasible implementations, the feature output layer 012 can be configured as a connection layer, so that the feature output layer 012 performs a connection operation on the first embedded feature and the second embedded feature to obtain a connection result, and outputs the connection result to the first intermediate layer 021, the second intermediate layer 022 and the third intermediate layer 023.

[0170] This disclosure does not limit the specific structure of the feature output layer 012. This disclosure can achieve the fusion of the first embedded feature and the second embedded feature by setting the connection operations in the feature output layer 012, further enhancing the media resource conversion attribution prediction model's ability to express data, thereby facilitating the accurate acquisition of media resource attribution results based fully on the information in the model's input data.

[0171] In some feasible embodiments, based on the media resource conversion attribution prediction model structure disclosed herein, a method for calculating the first and second probabilities of the media resource conversion attribution prediction model can be obtained. For example... Figure 4 The diagram illustrates a flowchart of step S30 in a media resource conversion attribution prediction model generation method according to an exemplary embodiment. The step of obtaining a first probability and a second probability based on media resource profile description information and user profile description information from the training samples includes:

[0172] In step S31, the first intermediate layer outputs the trigger conversion rate of the media resource based on the media resource profile description information and the user profile description information.

[0173] In step S32, the second intermediate layer outputs the user's trigger rate for media resources based on the media resource profile description information and the user profile description information.

[0174] In step S33, the third intermediate layer outputs the media resource non-trigger conversion rate based on the media resource profile description information and the user profile description information.

[0175] In step S34, the first output layer calculates the first product of the trigger conversion rate of the media resource output by the first intermediate layer and the user trigger rate of the media resource output by the second intermediate layer; and determines the first product as the first probability.

[0176] In step S35, the second output layer obtains an intermediate result based on the user trigger rate of media resources output by the second intermediate layer, calculates the second product of the intermediate result and the media resource non-trigger conversion rate output by the third intermediate layer, and determines the second product as the second probability.

[0177] In this disclosure, each factor used to obtain the relevant factors of the first probability and the second probability uses an independent intermediate layer output. The first output layer used to calculate the first probability is only connected to the relevant intermediate layer that outputs the factors related to the first probability. The second output layer used to calculate the second probability is only connected to the relevant intermediate layer that outputs the factors related to the second probability. It can be seen that for the media resource conversion attribution prediction model in this disclosure, both the setting of each logical layer of the model and the connection relationship between the logical layers can be adapted to the calculation methods of the first probability and the second probability to a large extent. Thus, the model complexity is reduced and the robustness of the model is improved by rationally setting the model structure.

[0178] In some feasible embodiments, based on the media resource conversion attribution prediction model structure disclosed herein, a method for determining the loss during training can be obtained. For example... Figure 5 The diagram illustrates a flowchart of step S40 in a media resource conversion attribution prediction model generation method according to an exemplary embodiment. The step of determining a loss value based on the first probability, the second probability, and the first label of the training samples includes:

[0179] In step S41, the first loss is calculated based on the first probability and the first label of the training sample.

[0180] In step S42, the second loss is calculated based on the second probability and the first label of the training sample.

[0181] In step S43, the loss value is calculated based on the first loss and the second loss.

[0182] In some feasible embodiments, the aforementioned loss values ​​can be obtained by directly summing the first loss and the second loss, or by weighted summation. If weighted summation is used, the weights corresponding to each loss can be preset or adaptively adjusted. Weighted summation facilitates the adjustment of parameters in the media resource conversion attribution prediction model, resulting in more accurate attribution results.

[0183] For example, the loss value can be expressed using the loss function loss = A * loss(scvr1) + B * loss(scvr0), where loss(scvr1) and loss(scvr0) represent the first loss and the second loss, respectively, and A and B can represent the weights of the loss, for example, A and B can both be 1. In some exemplary embodiments, loss(scvr1) and loss(scvr0) can be calculated using the cross-entropy loss function, and the specific calculation formulas are not limited in this disclosure.

[0184] In some feasible embodiments, based on the media resource conversion attribution prediction model structure disclosed herein, a method for determining the loss during training can be obtained. For example... Figure 6 The diagram illustrates a flowchart of step S40 in a media resource conversion attribution prediction model generation method according to an exemplary embodiment. The step of determining the loss value based on the first probability, the second probability, and the first label of the training samples includes:

[0185] In step S41-1, the first loss is calculated based on the first probability and the labels of the training samples.

[0186] In step S42-1, the second loss is calculated based on the second probability and the labels of the training samples.

[0187] In step S43-1, the third loss is calculated based on the user's trigger rate for media resources and the labels of the training samples.

[0188] Based on the media resource conversion attribution prediction model structure disclosed in this disclosure, the first probability scvr1 is output by the first output layer 031, i.e., scvr1 = pcvr1 * pctr; the second probability scvr0 is output by the second output layer 032, i.e., scvr0 = (1 - pctr) * pcvr0; and the user's trigger rate pctr for media resources is output by the second intermediate layer 022.

[0189] In step S44-1, the loss value is calculated based on the first loss, the second loss, and the third loss.

[0190] In some feasible embodiments, the aforementioned loss values ​​can be obtained by directly summing the first loss, the second loss, and the third loss, or by weighted summation. If weighted summation is used, the weights corresponding to each loss can be preset or adaptively adjusted. Weighted summation facilitates the adjustment of parameters in the media resource conversion attribution prediction model, resulting in more accurate attribution results.

[0191] For example, the loss value can be expressed using the loss function loss = A*loss(scvr1) + B*loss(scvr0) + C*loss(pctr), where loss(scvr1), loss(scvr0), and loss(pctr) represent the first loss, the second loss, and the third loss, respectively, and A, B, and C can represent the weights of the loss, for example, A, B, and C can all be set to 1. In some exemplary embodiments, loss(scvr1), loss(scvr0), and loss(pctr) can be calculated using the cross-entropy loss function, and the specific calculation formulas are not limited in this disclosure.

[0192] In some feasible implementations, step S50 in the above-mentioned method for generating a media resource conversion attribution prediction model can be implemented by adjusting the parameters of any fully connected layer among the first intermediate layer, the second intermediate layer, the third intermediate layer, and the information receiving layer. The embodiments of this disclosure do not limit the specific adjustment method.

[0193] This disclosure illustrates a method for generating a media resource conversion attribution prediction model. By designing the structure of the media resource conversion attribution prediction model, each intermediate layer in the model can independently output a factor used to obtain the attribution result. The first output layer used to calculate the first probability is only connected to the relevant intermediate layer that outputs the factor related to the first probability, and the second output layer used to calculate the second probability is only connected to the relevant intermediate layer that outputs the factor related to the second probability. This ensures that the setting of each logical layer and the connection relationships between logical layers in the media resource conversion attribution prediction model are largely compatible with the calculation methods of the first and second probabilities. Therefore, by rationally setting the model structure, the model complexity is reduced and the robustness of the model is improved. The construction and training process of the aforementioned media resource conversion attribution prediction model considers not only the case where the attribution result is user exposure conversion, but also the case where the attribution result is user-triggered conversion. Compared to related technologies that simply attribute media resources to either user-triggered conversion or media resource exposure conversion, resulting in low attribution accuracy, this embodiment can simultaneously output the probabilities of both user-triggered conversion and media resource exposure conversion, leading to more accurate attribution and significantly improved attribution accuracy. In one example, the media resource conversion attribution prediction model shown in this embodiment can improve the offline AUC by more than 0.7% compared to related technologies.

[0194] The following section uses a media resource conversion attribution prediction model based on the generated media resource conversion to further illustrate the practical application of the media resource conversion attribution prediction model provided in this disclosure.

[0195] Figure 7 This is a flowchart illustrating an application method of a media resource conversion attribution prediction model according to an exemplary embodiment, such as... Figure 7 As shown, a media resource conversion attribution prediction model is applied to... Figure 1 The following steps are described in the media resource server 120 shown.

[0196] In step S10-1, media resource profile description information and user profile description information are obtained.

[0197] This embodiment does not limit the specific content of the media resource profile description information and the user profile description information. They can be adapted to the input required by the media resource conversion attribution prediction model. For the media resource conversion attribution prediction model, please refer to the above text, and it will not be repeated here.

[0198] In step S20-1, the above media resource profile description information and user profile description information are input into the media resource conversion attribution prediction model to obtain the first probability, the second probability, and / or the media resource exposure conversion rate.

[0199] The media resource conversion attribution prediction model in this embodiment can output a first probability and a second probability. In step S20-1, either of the output probabilities can be selected according to actual needs, or the first and second output probabilities can be summed to obtain the media resource exposure conversion rate. The media resource exposure conversion rate can be used as a reference indicator for data processing in many media resource-related scenarios.

[0200] For example, the aforementioned media resource exposure conversion rate can be used as one of the parameters in the calculation of oCPM (Optimized Cost Per Mille), thus providing a reference for media resource owners when placing media resources. Under the oCPM model, when placing media resources on the media resource platform, media resource owners can set the target cost per click / conversion, while the media resource platform still charges based on the number of times the media resource is displayed.

[0201] For example, the aforementioned media resource exposure conversion rate can also be used as one of the parameters in the calculation of eCPM (effective cost per mille). Under the oCPM billing model, for each display opportunity, the media resource platform can calculate the eCPM and use the calculated eCPM as the bid for that display opportunity by that media resource. The media resource platform will then push the media resource with the highest bid to that display opportunity for display.

[0202] For example, the aforementioned media resource exposure conversion rate can also be used as one of the parameters in OCPC calculation, and the OCPC pricing model can be used to dynamically adjust prices to maximize the platform's eCPM revenue. In one example, based on the media resource conversion attribution prediction model provided in this disclosure embodiment, platform revenue can be increased by more than 10%.

[0203] The media resource conversion attribution prediction model application method shown in this embodiment can obtain relevant user profile description information and media resource profile description information, input the user profile description information and media resource profile description information into the media resource conversion attribution prediction model to obtain a first probability, a second probability and / or media resource exposure conversion rate, and then provide guidance for relevant media resource operation behavior based on the obtained first probability, second probability and / or media resource exposure conversion rate, such as participating in the calculation of performance parameters in relevant media resource operation behavior.

[0204] Figure 8 This is a flowchart illustrating a media resource push method according to an exemplary embodiment, such as... Figure 8 As shown, the media resource push method is applied to Figure 1 The following steps are described in the media resource server 120 shown.

[0205] In step S10-2, a media resource request is obtained, and the user profile description information in the media resource request is extracted.

[0206] This disclosure does not limit the specific content of the user profile description information, which can be adapted to the input required by the media resource conversion attribution prediction model. Please refer to the above for details on the media resource conversion attribution prediction model, which will not be repeated here. In some feasible implementations, the aforementioned user profile description information includes user-dimensional features and some or all of the user's media resource cross-features.

[0207] In step S20-2, a media resource set is obtained, and the media resource profile description information of each media resource in the media resource set is extracted.

[0208] In some feasible implementations, the aforementioned media resource profile description information includes media resource dimensional features, as well as some or all of the user's media resource cross-features.

[0209] In some possible implementations, the media resource set can also be filtered based on user profile description information, and the media resource profile description information of each media resource in the filtered media resource set can be extracted. The filtering action can perform preliminary screening of the media resource set, filtering out media resources that are determined to be unsuitable for delivery to users, reducing the computational load of media resource delivery, and improving the efficiency of media resource delivery.

[0210] In some possible implementations, at least one media resource favored by the user can be determined from the overall media resource set based on user preference information, thus forming a media resource set. Specifically, the aforementioned user preference information may be a subset of the aforementioned user profile description information, which may include, but is not limited to, at least one of the user's region, mobile phone model, age, and gender.

[0211] In step S30-2, for each media resource in the media resource set, the media resource profile description information and the user profile description information are input into the media resource conversion attribution prediction model to obtain the media resource exposure conversion rate of the media resource relative to the user; the media resource exposure conversion rate is the sum of the first probability and the second probability output by the media resource conversion attribution prediction model.

[0212] In step S40-2, the target media resource is determined based on the media resource exposure conversion rate corresponding to each media resource in the above media resource set.

[0213] In some possible implementations, the media resource with the highest exposure conversion rate can be identified as the target media resource.

[0214] In other possible implementations, the eCPM for each media resource can be calculated based on its exposure conversion rate, and the media resource with the highest eCPM can be identified as the target media resource. For example, in one implementation scenario, eCPM = media resource exposure rate * media resource main bid.

[0215] In step S50-2, the aforementioned target media resources are pushed.

[0216] The media resource push method shown in this embodiment can apply the media resource conversion attribution prediction model to the media resource push scenario. Based on the output of the media resource conversion attribution prediction model, the target media resources that are likely to generate returns are determined, and the target media resources are pushed to users. This not only meets the needs of users but also aligns with the interests of media resource owners, achieving a win-win result.

[0217] Figure 9 This is a block diagram illustrating a media resource conversion attribution prediction model generation apparatus according to an exemplary embodiment. (Refer to...) Figure 9 The device includes:

[0218] The media resource conversion attribution prediction model acquisition module 10 is configured to acquire the media resource conversion attribution prediction model. The media resource conversion attribution prediction model takes media resource profile description information and user profile description information as inputs and takes a first probability and a second probability as outputs. The first probability represents the probability that the attribution result is triggered by the user and the second probability represents the probability that the attribution result is converted by the media resource exposure.

[0219] The training sample set acquisition module 20 is configured to acquire a training sample set, which includes a first positive sample, a second positive sample, and a negative sample. Each of the first positive sample, the second positive sample, and the negative sample includes media resource profile description information, user profile description information, and a first tag. The first tag of the first positive sample represents that the attribution result is user-triggered conversion. The first tag of the second positive sample represents that the attribution result is media resource exposure conversion. The first tag of the negative sample represents that the media resource has not been converted.

[0220] The training prediction module 30 is configured to input the training samples into the media resource conversion attribution prediction model, so that the media resource conversion attribution prediction model obtains a first probability and a second probability based on the media resource profile description information and user profile description information of the training samples.

[0221] The loss value determination module 40 is configured to determine the loss value based on the first probability, the second probability and the first label of the training sample.

[0222] The training control module 50 is configured to adjust the parameters of the media resource conversion attribution prediction model based on the aforementioned loss value until the training termination condition is met.

[0223] In one exemplary embodiment, the media resource conversion attribution prediction model described above includes an information receiving layer, an intermediate layer, and an output layer;

[0224] The aforementioned intermediate layer is a flat structure consisting of a first intermediate layer, a second intermediate layer, and a third intermediate layer; the first intermediate layer, the second intermediate layer, and the third intermediate layer respectively correspond to the trigger conversion rate of the output media resources, the user's trigger rate for the media resources, and the non-trigger conversion rate of the media resources.

[0225] The aforementioned output layer is a flat structure composed of a first output layer and a second output layer; the first output layer and the second output layer respectively output the first probability and the second probability.

[0226] The first intermediate layer, the second intermediate layer, and the third intermediate layer are all connected to the information receiving layer; the first intermediate layer and the second intermediate layer are all connected to the first output layer; and the second intermediate layer and the third intermediate layer are all connected to the second output layer.

[0227] In an exemplary embodiment, the information receiving layer includes a feature data embedding layer and a feature output layer connected together, and the first intermediate layer, the second intermediate layer and the third intermediate layer are all connected to the feature output layer.

[0228] The aforementioned feature data embedding layer is used to embed sparse feature data in the aforementioned media resource profile description information and user profile description information to obtain the first embedded feature; the aforementioned sparse feature data includes media resource dimension features, user dimension features, and user-media resource cross features.

[0229] The aforementioned feature output layer is used to receive the aforementioned first embedded feature and second embedded feature, wherein the aforementioned second embedded feature is dense feature data from the aforementioned media resource profile description information and user profile description information.

[0230] In one exemplary implementation, the training prediction module described above is configured to perform:

[0231] The first intermediate layer outputs the trigger conversion rate of the media resource based on the media resource profile description information and the user profile description information.

[0232] The second intermediate layer outputs the user's trigger rate for media resources based on the media resource profile description information and the user profile description information.

[0233] The third intermediate layer above outputs the non-triggered conversion rate of media resources based on the media resource profile description information and the user profile description information.

[0234] The trigger conversion rate of the media resources output by the first intermediate layer is calculated by the first output layer and the first product of the user trigger rate of the media resources output by the second intermediate layer; the first product is determined as the first probability.

[0235] The second output layer obtains an intermediate result based on the user trigger rate of media resources output by the second intermediate layer, calculates the second product of the intermediate result and the non-trigger conversion rate of media resources output by the third intermediate layer, and determines the second product as the second probability.

[0236] In one exemplary embodiment, the loss value determination module includes:

[0237] The first loss calculation unit is configured to perform calculation of a first loss based on the first probability and the first label of the training sample.

[0238] The second loss calculation unit is configured to perform calculation of the second loss based on the second probability and the first label of the training sample.

[0239] The first loss totaling unit is configured to perform the calculation of the loss value based on the first loss and the second loss.

[0240] In one exemplary embodiment, the loss value determination module includes:

[0241] The first loss calculation unit is configured to perform calculation of a first loss based on the first probability and the first label of the training sample.

[0242] The second loss calculation unit is configured to perform calculation of the second loss based on the second probability and the first label of the training sample.

[0243] The third loss calculation unit is configured to perform a third loss calculation based on the user's trigger rate on the media resource and the second label of the training sample.

[0244] The second loss totaling unit calculates the aforementioned loss value based on the first loss, second loss, and third loss.

[0245] In one exemplary implementation, the aforementioned loss value is obtained by weighted summation of the individual losses it comprises.

[0246] Figure 10 This is a block diagram illustrating an application device for a media resource conversion attribution prediction model according to an exemplary embodiment. (Refer to...) Figure 10 The device includes:

[0247] The information acquisition module 10-1 is configured to acquire media resource profile description information and user profile description information.

[0248] Processing module 20-1 is configured to input the above media resource profile description information and user profile description information into the media resource conversion attribution prediction model to obtain a first probability, a second probability and / or media resource exposure conversion rate;

[0249] The media resource conversion attribution prediction model is obtained by using the media resource conversion attribution prediction model generation method of any of the above embodiments, and the media resource exposure conversion rate is the sum of the first probability and the second probability.

[0250] Figure 11 This is a block diagram illustrating a media resource delivery device according to an exemplary embodiment. (Refer to...) Figure 11 The device includes:

[0251] The media resource request acquisition module 10-2 is configured to execute the acquisition of media resource requests and extract user profile description information from the media resource requests.

[0252] The media resource set determination module 20-2 is configured to obtain the media resource set and extract the media resource profile description information of each media resource in the media resource set.

[0253] The media resource exposure conversion rate determination module 30-2 is configured to perform the following for each media resource in the above media resource set: input the media resource profile description information and the user profile description information into the media resource conversion attribution prediction model to obtain the media resource exposure conversion rate of the media resource relative to the user; the media resource exposure conversion rate is the sum of the first probability and the second probability output by the media resource conversion attribution prediction model.

[0254] The target media resource determination module 40-2 is configured to determine the target media resources based on the media resource exposure conversion rate corresponding to each media resource in the above media resource set.

[0255] Push module 50-2 is configured to push the aforementioned target media resources;

[0256] The aforementioned media resource conversion attribution prediction model is obtained using the media resource conversion attribution prediction model generation method of any of the above embodiments.

[0257] 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.

[0258] In an exemplary embodiment, an electronic device is also provided, including a processor; a memory for storing processor-executable instructions; wherein, when the processor is configured to execute the instructions stored in the memory, it implements the steps of the media resource conversion attribution prediction model generation method, the media resource conversion attribution prediction model application method, or the media resource push method provided in any of the above embodiments.

[0259] The electronic device can be a terminal, a server, or a similar computing device. Taking a server as an example... Figure 12 This is a block diagram illustrating an electronic device 1000 that executes a method for generating a media resource conversion attribution prediction model, a method for applying a media resource conversion attribution prediction model, or a method for pushing media resources, according to an exemplary embodiment. The electronic device 1000 can vary significantly due to different configurations or performance characteristics. It may include one or more central processing units (CPUs) 1010 (CPUs 1010 may include, but are not limited to, microprocessors (MCUs) or programmable logic devices (FPGAs), a memory 1030 for storing data, and one or more storage media 1020 (e.g., one or more mass storage devices) for storing application programs 1023 or data 1022. The memory 1030 and storage media 1020 may be temporary or persistent storage. The program stored in the storage media 1020 may include one or more modules, each module including a series of instruction operations on the electronic device. Furthermore, the CPU 1010 may be configured to communicate with the storage media 1020 and execute the series of instruction operations in the storage media 1020 on the electronic device 1000. Electronic device 1000 may also include one or more power supplies 1060, one or more wired or wireless network interfaces 1050, one or more input / output interfaces 1040, and / or one or more operating systems 1021, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0260] The input / output interface 1040 can be used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the electronic device 1000. In one example, the input / output interface 1040 includes a network interface controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In an exemplary embodiment, the input / output interface 100 can be a radio frequency (RF) module for wireless communication with the Internet.

[0261] Those skilled in the art will understand that Figure 12 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, the electronic device 1000 may also include components that are more... Figure 12 The more or fewer components shown, or having the same Figure 12 The different configurations shown.

[0262] In an exemplary embodiment, a storage medium is also provided, which, when the instructions in the storage medium are executed by the processor of an electronic device, enables the electronic device to perform the steps of any of the media resource conversion attribution prediction model generation method, media resource conversion attribution prediction model application method, or media resource push method in the above embodiments.

[0263] In an exemplary embodiment, a computer program product is also provided, comprising computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the media resource conversion attribution prediction model generation method, media resource conversion attribution prediction model application method, or media resource push method provided in any of the above embodiments.

[0264] 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.

[0265] Other embodiments of this disclosure 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 disclosure that follow the general principles of this disclosure 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 disclosure are indicated by the following claims.

[0266] It should be understood that this disclosure is not limited to the precise structures 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 disclosure is limited only by the appended claims.

Claims

1. A method for generating a media resource conversion attribution prediction model, characterized in that, include: A media resource conversion attribution prediction model is obtained, which outputs the probability of media resource exposure conversion as an attribution result and the probability of user-triggered conversion as an attribution result. The media resource conversion attribution prediction model takes media resource profile description information and user profile description information as inputs, and outputs a first probability and a second probability. The first probability represents the probability that the attribution result is user-triggered conversion, and the second probability represents the probability that the attribution result is media resource exposure conversion. Obtain a training sample set, which includes a first positive sample, a second positive sample, and a negative sample. The first positive sample, the second positive sample, and the negative sample all include media resource profile description information, user profile description information, and a first tag. The first tag of the first positive sample represents that the attribution result is user-triggered conversion. The first tag of the second positive sample represents that the attribution result is media resource exposure conversion. The first tag of the negative sample represents that the media resource has not been converted. The training samples are input into the media resource conversion attribution prediction model so that the media resource conversion attribution prediction model obtains a first probability and a second probability based on the media resource profile description information and user profile description information of the training samples. The loss value is determined based on the first probability, the second probability, and the first label of the training sample; The parameters of the media resource conversion attribution prediction model are adjusted based on the loss value until the training stopping condition is met.

2. The method for generating a media resource conversion attribution prediction model according to claim 1, characterized in that: The media resource conversion attribution prediction model includes an information receiving layer, an intermediate layer, and an output layer; The intermediate layer is a flat structure consisting of a first intermediate layer, a second intermediate layer, and a third intermediate layer; the first intermediate layer, the second intermediate layer, and the third intermediate layer respectively output the trigger conversion rate of media resources, the user's trigger rate of media resources, and the non-trigger conversion rate of media resources. The output layer is a flat structure composed of a first output layer and a second output layer; the first output layer and the second output layer respectively output a first probability and a second probability. The first intermediate layer, the second intermediate layer, and the third intermediate layer are all connected to the information receiving layer; the first intermediate layer and the second intermediate layer are both connected to the first output layer; and the second intermediate layer and the third intermediate layer are both connected to the second output layer.

3. The method for generating a media resource conversion attribution prediction model according to claim 2, characterized in that, The information receiving layer includes a feature data embedding layer and a feature output layer connected together, and the first intermediate layer, the second intermediate layer and the third intermediate layer are all connected to the feature output layer; The feature data embedding layer is used to embed sparse feature data in the media resource profile description information and user profile description information to obtain the first embedded feature; The sparse feature data includes media resource dimension features, user dimension features, and user-media resource cross features; The feature output layer is used to receive the first embedded feature and the second embedded feature, wherein the second embedded feature is dense feature data in the media resource profile description information and the user profile description information.

4. The method for generating a media resource conversion attribution prediction model according to claim 2 or 3, characterized in that, The step of obtaining the first probability and the second probability based on the media resource profile description information and user profile description information of the training samples includes: The first intermediate layer outputs the trigger conversion rate of the media resource based on the media resource profile description information and the user profile description information; The second intermediate layer outputs the user's trigger rate for media resources based on the media resource profile description information and the user profile description information; The third intermediate layer outputs the non-triggered conversion rate of media resources based on the media resource profile description information and the user profile description information; The first output layer calculates the first product of the trigger conversion rate of the media resource output by the first intermediate layer and the user trigger rate of the media resource output by the second intermediate layer; the first product is determined as the first probability. The second output layer obtains an intermediate result based on the user trigger rate of media resources output by the second intermediate layer, calculates the second product of the intermediate result and the non-trigger conversion rate of media resources output by the third intermediate layer, and determines the second product as the second probability.

5. The method for generating a media resource conversion attribution prediction model according to claim 4, characterized in that, Determining the loss value based on the first probability, the second probability, and the first label of the training sample includes: Calculate the first loss based on the first probability and the first label of the training sample; The second loss is calculated based on the second probability and the first label of the training sample; The loss value is calculated based on the first loss and the second loss.

6. The method for generating a media resource conversion attribution prediction model according to claim 4, characterized in that, The training samples in the training sample set also include a second label, which is used to characterize whether a user-triggered action exists. Correspondingly, determining the loss value based on the first probability, the second probability, and the first label of the training samples includes: Calculate the first loss based on the first probability and the first label of the training sample; The second loss is calculated based on the second probability and the first label of the training sample; The third loss is calculated based on the user's trigger rate for media resources and the second label of the training samples; The loss value is calculated based on the first loss, the second loss, and the third loss.

7. The method for generating a media resource conversion attribution prediction model according to claim 5 or 6, characterized in that, The loss value is obtained by weighted summation of the individual losses it includes.

8. A method for applying a media resource conversion attribution prediction model, characterized in that, include: Obtain media resource profile description information and user profile description information; Input the media resource profile description information and user profile description information into the media resource conversion attribution prediction model to obtain the first probability, the second probability and / or the media resource exposure conversion rate. The media resource conversion attribution prediction model is obtained using the media resource conversion attribution prediction model generation method as described in any one of claims 1-7, and the media resource exposure conversion rate is the sum of the first probability and the second probability.

9. A method for pushing media resources, characterized in that, include: Obtain media resource requests and extract user profile description information from the media resource requests; Obtain a media resource set and extract the media resource profile description information for each media resource in the media resource set; For each media resource in the media resource set, the media resource profile description information and the user profile description information are input into the media resource conversion attribution prediction model to obtain the media resource exposure conversion rate of the media resource relative to the user; the media resource exposure conversion rate is the sum of the first probability and the second probability output by the media resource conversion attribution prediction model. The target media resources are determined based on the media resource exposure conversion rate of each media resource in the media resource set. Push the target media resources; The media resource conversion attribution prediction model is obtained using the media resource conversion attribution prediction model generation method as described in any one of claims 1-7.

10. A media resource conversion attribution prediction model generation device, characterized in that, include: The media resource conversion attribution prediction model acquisition module is configured to acquire the media resource conversion attribution prediction model. The media resource conversion attribution prediction model is used to output the probability of media resource exposure conversion as an attribution result and the probability of user-triggered conversion as an attribution result. The media resource conversion attribution prediction model takes media resource profile description information and user profile description information as inputs and outputs a first probability and a second probability. The first probability represents the probability that the attribution result is user-triggered conversion, and the second probability represents the probability that the attribution result is media resource exposure conversion. The training sample set acquisition module is configured to acquire a training sample set, which includes a first positive sample, a second positive sample, and a negative sample. Each of the first positive sample, the second positive sample, and the negative sample includes media resource profile description information, user profile description information, and a first tag. The first tag of the first positive sample represents that the attribution result is user-triggered conversion. The first tag of the second positive sample represents that the attribution result is media resource exposure conversion. The first tag of the negative sample represents that the media resource has not been converted. The training prediction module is configured to input the training samples into the media resource conversion attribution prediction model, so that the media resource conversion attribution prediction model obtains a first probability and a second probability based on the media resource profile description information and user profile description information of the training samples. The loss value determination module is configured to determine a loss value based on the first probability, the second probability, and the first label of the training sample; The training control module is configured to adjust the parameters of the media resource conversion attribution prediction model based on the loss value until the training stop condition is met.

11. The media resource conversion attribution prediction model generation device according to claim 10, characterized in that: The media resource conversion attribution prediction model includes an information receiving layer, an intermediate layer, and an output layer; The intermediate layer is a flat structure consisting of a first intermediate layer, a second intermediate layer, and a third intermediate layer; the first intermediate layer, the second intermediate layer, and the third intermediate layer respectively output the trigger conversion rate of media resources, the user's trigger rate of media resources, and the non-trigger conversion rate of media resources. The output layer is a flat structure composed of a first output layer and a second output layer; the first output layer and the second output layer respectively output a first probability and a second probability. The first intermediate layer, the second intermediate layer, and the third intermediate layer are all connected to the information receiving layer; the first intermediate layer and the second intermediate layer are both connected to the first output layer; and the second intermediate layer and the third intermediate layer are both connected to the second output layer.

12. The media resource conversion attribution prediction model generation device according to claim 11, characterized in that, The information receiving layer includes a feature data embedding layer and a feature output layer connected together, and the first intermediate layer, the second intermediate layer and the third intermediate layer are all connected to the feature output layer; The feature data embedding layer is used to embed sparse feature data in the media resource profile description information and user profile description information to obtain the first embedded feature; The sparse feature data includes media resource dimension features, user dimension features, and user-media resource cross features; The feature output layer is used to receive the first embedded feature and the second embedded feature, wherein the second embedded feature is dense feature data in the media resource profile description information and the user profile description information.

13. The media resource conversion attribution prediction model generation device according to claim 11 or 12, characterized in that, The training prediction module is configured to execute: The first intermediate layer outputs the trigger conversion rate of the media resource based on the media resource profile description information and the user profile description information; The second intermediate layer outputs the user's trigger rate for media resources based on the media resource profile description information and the user profile description information; The third intermediate layer outputs the non-triggered conversion rate of media resources based on the media resource profile description information and the user profile description information; The first output layer calculates the first product of the trigger conversion rate of the media resources output by the first intermediate layer and the user trigger rate of the media resources output by the second intermediate layer; The first product is determined as the first probability; The second output layer obtains an intermediate result based on the user trigger rate of media resources output by the second intermediate layer, calculates the second product of the intermediate result and the non-trigger conversion rate of media resources output by the third intermediate layer, and determines the second product as the second probability.

14. The media resource conversion attribution prediction model generation device according to claim 13, characterized in that, The loss value determination module includes: The first loss calculation unit is configured to perform calculation of a first loss based on the first probability and the first label of the training sample; The second loss calculation unit is configured to perform a second loss calculation based on the second probability and the first label of the training sample; The first loss totaling unit is configured to perform the calculation of the loss value based on the first loss and the second loss.

15. The media resource conversion attribution prediction model generation device according to claim 13, characterized in that, The loss value determination module includes: The first loss calculation unit is configured to perform calculation of a first loss based on the first probability and the first label of the training sample; The second loss calculation unit is configured to perform a second loss calculation based on the second probability and the first label of the training sample; The third loss calculation unit is configured to perform a third loss calculation based on the user's trigger rate on the media resource and the second label of the training sample; The second loss totaling unit calculates the loss value based on the first loss, the second loss, and the third loss.

16. The media resource conversion attribution prediction model generation device according to claim 14 or 15, characterized in that, The loss value is obtained by weighted summation of the individual losses it includes.

17. A media resource conversion attribution prediction model application device, characterized in that, include: The information acquisition module is configured to acquire media resource profile description information and user profile description information; The processing module is configured to input the media resource profile description information and user profile description information into the media resource conversion attribution prediction model to obtain a first probability, a second probability, and / or the media resource exposure conversion rate. The media resource conversion attribution prediction model is obtained using the media resource conversion attribution prediction model generation method as described in any one of claims 1-7, and the media resource exposure conversion rate is the sum of the first probability and the second probability.

18. A media resource delivery device, characterized in that, include: The media resource request acquisition module is configured to acquire media resource requests and extract user profile description information from the media resource requests; The media resource set determination module is configured to acquire the media resource set and extract the media resource profile description information of each media resource in the media resource set; The media resource exposure conversion rate determination module is configured to perform, for each media resource in the media resource set, input the media resource profile description information and the user profile description information into the media resource conversion attribution prediction model to obtain the media resource exposure conversion rate of the media resource relative to the user; the media resource exposure conversion rate is the sum of the first probability and the second probability output by the media resource conversion attribution prediction model. The target media resource determination module is configured to determine the target media resources based on the media resource exposure conversion rate corresponding to each media resource in the media resource set; The push module is configured to push the target media resource; The media resource conversion attribution prediction model is obtained using the media resource conversion attribution prediction model generation method as described in any one of claims 1-7.

19. 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 media resource conversion attribution prediction model generation method as described in any one of claims 1 to 7, the media resource conversion attribution prediction model application method as described in claim 8, or the media resource push method as described in claim 9.

20. A storage medium, wherein instructions in the storage medium, when executed by a processor of an electronic device, enable the electronic device to perform the media resource conversion attribution prediction model generation method as described in any one of claims 1 to 7, the media resource conversion attribution prediction model application method as described in claim 8, or the media resource push method as described in claim 9.

Images