Media data processing method and device, computer equipment and storage medium

Abstract

The invention relates to a media data processing method and device, computer equipment, a storage medium and a computer program product. The method can be applied to an artificial intelligence scene and a media recommendation scene in smart traffic, and comprises the following steps: analyzing media characteristic data from media form data of media to be recommended; splitting each piece of media characteristic data into different morphological characteristic parameters; mapping the morphological characteristic parameters of the media characteristic data into morphological characteristics of a preset type; performing feature intersection on the morphological features and the media traffic features to obtain first intersection features; performing feature processing on the first cross feature to obtain a first target feature, performing feature processing on the first cross feature, the morphological feature, the media feature of the to-be-recommended media and the object feature of the target object to obtain a second target feature, and determining the interaction rate of the to-be-recommended media according to the first target feature and the second target feature. By adopting the method, the pre-estimation accuracy of the media interaction rate can be effectively improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and particularly to a method, apparatus, computer device, storage medium, and computer program product for processing media data. Background Art

[0002] With the continuous development of Internet technology, users can enjoy various digital product services through the Internet. In order to promote other products, service providers usually recommend media recommendation information of products that users may be interested in during the process of users enjoying digital product services, which is conducive to improving the interaction rate of the product. However, as a factor affecting the interaction rate, the traditional solution usually directly uses the media form as the input feature of the model to estimate the interaction rate. With the development and change of the media form, it may affect the accuracy of the model estimation. Summary of the Invention

[0003] Based on this, in view of the above technical problems, it is necessary to provide a method, apparatus, computer device, computer-readable storage medium, and computer program product for processing media data, which can effectively improve the estimation accuracy of the media interaction rate.

[0004] In a first aspect, the present application provides a method for processing media data, the method comprising:

[0005] Parsing at least two media characteristic data from the media form data of the media to be recommended;

[0006] Respectively splitting each of the media characteristic data into different form characteristic parameters;

[0007] Successively mapping the form characteristic parameters of each of the media characteristic data into form features of a preset type;

[0008] Performing feature crossing on the form features and media traffic features to obtain first crossing features;

[0009] Through a media estimation model, performing feature processing on the first crossing features to obtain first target features, performing feature processing on the first crossing features, the form features, the media features of the media to be recommended, and the object features of the target object to obtain second target features, and determining the interaction rate of the media to be recommended according to the first target features and the second target features.

[0010] In a second aspect, the present application further provides a media data processing apparatus, the apparatus comprising:

[0011] A parsing module, configured to parse at least two media characteristic data from the media form data of the media to be recommended;

[0012] A splitting module, configured to split each of the media characteristic data into different morphological characteristic parameters respectively;

[0013] A mapping module, configured to map the morphological characteristic parameters of each of the media characteristic data into morphological features of a preset type in sequence;

[0014] An intersection module, configured to perform feature intersection on the morphological features and media traffic features to obtain first intersection features;

[0015] An estimation module, configured to perform feature processing on the first intersection features through a media estimation model to obtain first target features, perform feature processing on the first intersection features, the morphological features, the media features of the media to be recommended, and the object features of the target object to obtain second target features, and determine the interaction rate of the media to be recommended according to the first target features and the second target features.

[0016] In one embodiment, the number of the media to be recommended is at least two; the apparatus further includes:

[0017] A receiving module, configured to receive a media recommendation request sent by a client;

[0018] A first obtaining module, configured to obtain at least two media to be recommended based on the media recommendation request;

[0019] A storage module, configured to save at least two of the media to be recommended in a media candidate queue;

[0020] A searching module, configured to search for matching media morphological data for each of the media to be recommended in the media candidate queue in sequence.

[0021] In one embodiment, the searching module is further configured to obtain media morphological selection conditions configured on a media morphological configuration platform; select media morphological data matching each of the media to be recommended in the media candidate queue according to the media morphological selection conditions; wherein, the media morphological configuration platform is used to create various types of candidate media morphological data.

[0022] In one embodiment, the apparatus further includes:

[0023] The first obtaining module is further configured to obtain media morphological description data of an operation object on the media morphological configuration platform;

[0024] A first generating module, configured to generate candidate media morphological data in a graphical or graphic-text form according to the media morphological description data, the media traffic features, and the object features of the target object.

[0025] The storage module is further configured to store the candidate media form data in a database in response to a data storage operation triggered on the media form configuration platform;

[0026] The first acquisition module is further configured to acquire at least two media to be recommended corresponding to the media recommendation request from the database.

[0027] In one embodiment, the mapping module is further configured to use the parameter names in the form feature parameters of each media feature as keys and the parameter values in the form feature parameters of each media feature as values to obtain a plurality of different key-value pairs; obtain the mapping relationship between the form feature parameters and the form features of the preset type; and write each key-value pair into a form feature file in a target format according to the mapping relationship to obtain form features of the preset type.

[0028] In one embodiment, the apparatus further includes:

[0029] A second acquisition module, configured to acquire media traffic features; the media traffic features include at least one of a media site identifier, a scenario identifier, or a media display position identifier;

[0030] The cross module is further configured to perform a hashing process on the form features to obtain form feature values;

[0031] Perform a hashing process on the media traffic features to obtain media traffic feature values; perform feature crossing on the form feature values and the media traffic feature values to obtain a first cross feature.

[0032] In one embodiment, the apparatus further includes:

[0033] A third acquisition module, configured to acquire the media features of the media to be recommended and the object features of the target object;

[0034] The estimation module is further configured to encode the first cross feature through a first network branch in the media estimation model to obtain a first embedding representation, and perform feature processing on the first embedding representation to obtain a first target feature; encode the first cross feature, the form features, the media features of the media to be recommended, and the object features of the target object through a second network branch in the media estimation model to obtain a second embedding representation, and perform feature processing on the second embedding representation to obtain a second target feature.

[0035] In one embodiment, the media estimation model is an online media estimation model; the apparatus further includes:

[0036] An extraction module, configured to input the media form data of the media to be recommended into an offline network model, so that the offline network model extracts an embedded representation of the media form data to obtain a third embedded representation;

[0037] The cross module is further configured to perform feature crossing on the third embedded representation as a form feature and a media traffic feature to obtain a second cross feature;

[0038] The prediction module is further configured to perform feature processing on the second cross feature through the online media prediction model to obtain a third target feature, perform feature processing on the second cross feature, the third embedded representation, the media feature of the media to be recommended, and the object feature of the target object to obtain a fourth target feature, and determine the interaction rate of the media to be recommended according to the third target feature and the fourth target feature.

[0039] In one embodiment, the apparatus further includes:

[0040] The parsing module is further configured to parse the media form data of the sample media to obtain at least two training media characteristic data;

[0041] The splitting module is further configured to split each of the training media characteristic data into different training form characteristic parameters;

[0042] The mapping module is further configured to sequentially map each of the training form characteristic parameters into a training form feature of the preset type;

[0043] The cross module is further configured to perform feature crossing on the training form feature and the media traffic feature corresponding to the sample media to obtain a third cross feature;

[0044] The prediction module is further configured to perform feature processing on the third cross feature through a media prediction model to obtain a third target feature, perform feature processing on the third cross feature, the training form feature, the media feature of the sample media, and the object feature of the sample object to obtain a fourth target feature, and determine the interaction rate of the sample media according to the third target feature and the fourth target feature;

[0045] An optimization module, configured to optimize the media prediction model according to the loss value between the interaction rate and the interaction label of the sample media.

[0046] In one embodiment, the interaction rate includes a conversion rate and a click-through rate, and the number of media to be recommended is at least two; the apparatus further includes:

[0047] A fourth acquisition module, configured to acquire a media recommendation resource value proposed by a product party for recommending the product of the product party; in at least two of the to-be-recommended media, select a second media according to the media recommendation resource value, the conversion rate, and the click-through rate.

[0048] A first push module, configured to push the second media and media form data corresponding to the second media to the target object, so as to display the second media on the application page of the target object according to the media form data.

[0049] In one embodiment, the interaction rate includes a conversion rate and a click-through rate, and the number of the to-be-recommended media is at least two; the apparatus further includes:

[0050] A selection module, configured to select a first media whose interaction rate meets an interaction condition from at least two of the to-be-recommended media.

[0051] A second push module, configured to push the first media and media form data corresponding to the first media to the target object, so as to display the first media on the application page of the target object according to the media form data.

[0052] In a third aspect, the present application further provides a computer device, where the computer device includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the steps of the media data processing method are implemented.

[0053] In a fourth aspect, the present application further provides a computer-readable storage medium, where the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the media data processing method are implemented.

[0054] In a fifth aspect, the present application further provides a computer program product, where the computer program product includes a computer program, and when the computer program is executed by a processor, the steps of the media data processing method are implemented.

[0055] The above media data processing method, device, computer equipment, storage medium, and computer program product, before estimating the interaction rate using the media prediction model, first parse at least two media characteristic data from the media form data of the media to be recommended, split each media characteristic data into different form characteristic parameters, and sequentially map the form characteristic parameters of each media characteristic data into form features of a preset type. Thus, no matter how the media form changes, it can be transformed into form features that can represent the media form. Therefore, when estimating the interaction rate using the media prediction model, not only can the perception of the new media form be realized, but also the estimation accuracy can be effectively improved. In addition, feature crossing is performed on the form features and media traffic features. Therefore, no matter what the source of the media traffic is, by using the media prediction model to process the first cross feature to obtain the first target feature, and processing the first cross feature, form features, media features of the media to be recommended, and object features of the target object to obtain the second target feature, the interaction rate of the media to be recommended can be accurately determined according to the first target feature and the second target feature, further improving the accuracy of the interaction rate estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is an application environment diagram of the media data processing method in an embodiment;

[0057] Figure 2 It is a flowchart of the media data processing method in an embodiment;

[0058] Figure 3 It is a schematic diagram of a media form in an embodiment;

[0059] Figure 4 It is a schematic diagram of constructing form features in an embodiment;

[0060] Figure 5 It is a schematic diagram of constructing form features and generating feature values in an embodiment;

[0061] Figure 6 It is a schematic diagram of a media prediction model in an embodiment;

[0062] Figure 7 It is a flowchart of training the media prediction model in an embodiment;

[0063] Figure 8 It is a schematic diagram of a media data processing architecture in an embodiment;

[0064] Figure 9 It is a schematic diagram of the advertisement form in the advertisement display stage in an embodiment;

[0065] Figure 10Schematic diagram of the advertisement form in the advertisement display stage in another embodiment;

[0066] Figure 11 Schematic diagram of the advertisement form in the advertisement playing stage in one embodiment;

[0067] Figure 12 Schematic diagram of the advertisement form in the advertisement interaction stage in one embodiment;

[0068] Figure 13 Schematic diagram of the advertisement form in the advertisement conversion stage in one embodiment;

[0069] Figure 14 Schematic diagram of converting the advertisement form into advertisement features in one embodiment;

[0070] Figure 15 Schematic diagram of converting the advertisement form into advertisement features in another embodiment;

[0071] Figure 16 Structural block diagram of the media data processing device in one embodiment;

[0072] Figure 17 Internal structure diagram of the computer device in one embodiment. Specific implementation manners

[0073] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0074] It should be noted that in the following descriptions, the terms "first", "second" and "third" only distinguish similar objects, and do not represent a specific order for the objects. It can be understood that "first", "second" and "third" can be interchanged with a specific order or sequence when permitted, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein.

[0075] Before describing the embodiments of the present application, the technologies involved in the present application are introduced as follows:

[0076] Artificial Intelligence (AI) is a theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a way similar to human intelligence. Artificial intelligence also studies the design principles and implementation methods of various intelligent machines, enabling machines to have the functions of perception, reasoning, and decision-making.

[0077] Artificial intelligence technology is an interdisciplinary subject with a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, pre-trained model technology, operation / interaction systems, mechatronics, etc. Among them, the pre-trained model, also known as the large model or the foundation model, can be widely applied to downstream tasks in various directions of artificial intelligence after fine-tuning. The software technologies of artificial intelligence mainly include several major directions such as computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0078] The media data processing method provided in the embodiments of this application can be applied to an application environment as Figure 1 shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or placed in the cloud or other network servers.

[0079] The server 104 parses at least two media characteristic data from the media form data of the media to be recommended; splits each media characteristic data into different form characteristic parameters; maps the form characteristic parameters of each media characteristic data into form features of a preset type in sequence; performs feature crossing on the form features and media traffic features to obtain a first crossed feature; through a media prediction model, performs feature processing on the first crossed feature to obtain a first target feature, performs feature processing on the first crossed feature, form features, media features of the media to be recommended, and object features of the target object to obtain a second target feature, and determines the interaction rate of the media to be recommended according to the first target feature and the second target feature. The interaction rate can include the conversion rate and the click-through rate. Therefore, the server 104 can obtain the media recommendation resource value proposed by the product party for recommending the product of the product party; among at least two media to be recommended, select the second media according to the media recommendation resource value, conversion rate, and click-through rate; push the second media and the corresponding media form data to the terminal 102 held by the target object to display the second media on the application page of the target object according to the media form data.

[0080] Among them, the terminal 102 can be a smart phone, a tablet computer, a notebook computer, a desktop computer, an Internet of Things device, and a portable wearable device. The Internet of Things device can be a smart speaker, a smart TV, a smart air conditioner, a smart vehicle device, etc. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, etc.

[0081] The server 104 can be an independent physical server or a service node in a blockchain system. A peer-to-peer (Peer To Peer) network is formed among the service nodes in the blockchain system. The peer-to-peer protocol is an application layer protocol running on top of the Transmission Control Protocol (TCP). In addition, the server 104 can also be a server cluster composed of multiple physical servers, and can be a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, Content Delivery Network (CDN), and big data and artificial intelligence platforms.

[0082] In one embodiment, as Figure 2 shown, a media data processing method is provided. This method can be executed by the Figure 1 server or terminal in Figure 1 , or executed jointly by the server and the terminal. Taking the execution of this method by the

[0083] server in

[0084] as an example, the method includes the following steps:

[0083] S202, parse at least two media characteristic data from the media form data of the media to be recommended.

[0084] Among them, the media to be recommended can be media for promoting a product that needs to be pushed to a target object. At this time, the presentation form of the media to be recommended can be one or more (i.e., two or more than two data) combinations of videos, images, animations, and special effect texts about the product. The product can be a smart device, a car, a game, an application software, etc. that needs to be promoted. In addition, the media to be recommended can also be a media task for promoting a product that needs to be pushed to a target object.

[0085] The media form data can refer to data used to represent the form presented by the media to be recommended, and can be graphical or graphic-text data. For example, each frame of media form data can be graphical or graphic-text data used to represent the form presented by the media to be recommended, as Figure 3 shown. Among them, the media form data in different stages can be different. For example, the media form data in the display stage, the playing stage, the interaction stage, and the conversion stage of the media to be recommended can be different.

[0086] The media characteristic data may include data for representing the interactive characteristics of the media to be recommended, such as swipe click, shake click, and card click, etc.; in addition, it may also include text descriptions for promoting the media to be recommended, such as introduction information and incentive information.

[0087] In one embodiment, the server may determine an interaction method that meets the interaction conditions in the media form data of the media to be recommended, and generate at least two media characteristic data that conform to the interaction method. Alternatively, the server may generate at least two media characteristic data according to the interaction method corresponding to the media form data of the media to be recommended. Among them, the interaction method corresponding to the media form data may be set on the media form configuration platform.

[0088] In one embodiment, the number of media to be recommended is at least two (i.e., multiple). Therefore, before S202, the server may also receive a media recommendation request sent by the client; obtain at least two media to be recommended based on the media recommendation request; save the at least two media to be recommended in the media candidate queue; and sequentially find matching media form data for each media to be recommended in the media candidate queue.

[0089] Among them, the client may be a video client, a game client, a browser, a map, and other life-related clients (such as life sharing clients, shopping clients, and food delivery clients, etc.).

[0090] The media recommendation request may be a network request for media recommendation to a target object. The media recommendation request may be a request sent by the client according to the configured trigger timing, or a request sent in response to the trigger operation of the target object.

[0091] The target object may be a user who needs to push the media to be recommended during application interactions in scenarios such as watching videos, playing games, browsing the web, etc. Among them, the above-mentioned videos may include on-demand videos and live videos, and the live videos may be live sports videos, concert videos, and videos in the live broadcast room.

[0092] For example, when a user is watching a video, in order to increase the click-through rate and conversion rate of a product that a certain user may be interested in, an advertisement for the product may be pushed to the user. Among them, in different application scenarios, the conversion rate may be the download rate, purchase rate, or usage rate, etc.

[0093] Specifically, after receiving a media recommendation request sent by a client, the server obtains at least two media to be recommended based on the media recommendation request in a media database, then sorts the at least two media to be recommended according to interest scores, then saves the sorted media to be recommended in a media candidate queue, and then sequentially finds matching media form data for each media to be recommended in the media candidate queue.

[0094] In one embodiment, the step of sequentially finding matching media form data for each media to be recommended in the media candidate queue may specifically include: the server obtains media form selection conditions configured on a media form configuration platform; according to the media form selection conditions, selects media form data matching each media to be recommended in the candidate media form data.

[0095] Among them, the media form configuration platform is used to create various types of candidate media form data, and the candidate media form data may be data used to represent the media form presented by the media to be recommended, and may be graphical or graphic-text form data. For example, each frame of candidate media form data may be graphical or graphic-text form data used to represent the media form.

[0096] The media form selection conditions may include media form selection rules and / or strategies. For example, for the advertising scenario, considering that form strategies such as the display duration, interaction method, landing page style, and user reward mode of the advertisement affect the click-through rate and conversion rate of the advertisement, the media form data is selected according to the above form strategies.

[0097] For example, for the advertising scenario, by adjusting the display duration, interaction method, landing page style, and user reward mode of the advertisement, etc., different people have different experiences of seeing different advertisements at different times, so as to improve the click-through rate of the advertisement, and further improve the expected revenue per thousand impressions (eCPM) and traffic revenue of the advertisement.

[0098] In one embodiment, the media form configuration platform obtains media form description data of an operation object; generates graphical or graphic-text form candidate media form data according to the media form description data, media traffic characteristics, and object characteristics of the target object; in response to a data storage operation triggered on the media form configuration platform, stores the candidate media form data in a database; therefore, when obtaining media to be recommended, at least two media to be recommended corresponding to the media recommendation request can be obtained from the database.

[0099] Among them, the operation object can input requirement data on the dialogue sub-page of the media form configuration platform. The requirement data is media form description data. The media form configuration platform can use a large model to generate candidate media form data in graphical or graphic-text form according to the media form description data, media traffic characteristics, and object characteristics of the target object.

[0100] In addition, the operation object can also create media form description data on the media form configuration platform based on the media traffic characteristics and object characteristics of the target object by using the creation tools on the media form configuration platform.

[0101] S204, split each media characteristic data into different form characteristic parameters respectively.

[0102] Among them, the form characteristic parameters can be parameters obtained by splitting the media characteristic data and used to represent the small-particle interaction characteristics of the media to be recommended.

[0103] In one embodiment, the server can obtain a pre-configured data splitting strategy and directly split each media characteristic data into different form characteristic parameters according to the data splitting strategy. For example, if the media characteristic data is sliding click, the media characteristic data can be split into trigger area, sliding distance, sliding angle, etc. according to the data splitting strategy of sliding click.

[0104] In another embodiment, the server can obtain the parameter mapping relationship corresponding to each media characteristic data that is pre-configured, and obtain the form characteristic parameters matching each media characteristic data according to the parameter mapping relationship, so as to realize the splitting of each media characteristic data and obtain the corresponding form characteristic parameters.

[0105] S206, map the form characteristic parameters of each media characteristic data into form characteristics of a preset type in sequence.

[0106] Among them, there are a fixed number of types (i.e., preset types) of form characteristics, that is, there are a fixed number of form characteristics, including: component display timing, component style, media display time, playback attribute, click mode, click area, retention prompt (such as retention pop-up window), landing page style, download mode, landing page pull-up method, and incentive mode, etc. The above form characteristics are only examples rather than an exhaustive list, and can be dynamically added or deleted according to the actual situation.

[0107] It should be noted that the number of form characteristic parameters is greater than the number of media characteristic data, and the number of media characteristic data is greater than the number of form characteristics, that is: the relationship between form characteristic parameters and media characteristic data is many-to-one, and the relationship between media characteristic data and form characteristics is many-to-one.

[0108] In one embodiment, the server may map the morphological feature parameters into morphological features of a preset type according to the mapping relationship between the morphological feature parameters and the morphological features.

[0109] Among them, the mapping relationship may be a preset association relationship for converting morphological feature parameters into morphological features. For example, according to mapping relationship 1, morphological feature parameter A and morphological feature parameter B are converted into morphological feature X; and for another example, according to mapping relationship 2, morphological feature parameter C is converted into morphological feature Y, as Figure 4 shown.

[0110] Specifically, the server uses the parameter names in the morphological characteristic parameters of each media characteristic as keys and the parameter values in the morphological characteristic parameters of each media characteristic as values to obtain a plurality of different key-value pairs; obtains the mapping relationship between the morphological feature parameters and the morphological features of the preset type; and writes each key-value pair into the morphological feature file in the target format according to the mapping relationship to obtain the morphological features of the preset type.

[0111] Among them, the target format may be a structured data storage format, such as the Protocol Buffers (PB) format, which is used for serializing or serializing structured data. The morphological feature file may be used to store serialized or serialized data or features.

[0112] In one embodiment, each different morphological feature may be a feature data set. Therefore, the server may write each key-value pair into the corresponding morphological feature of the preset type in the morphological feature file according to the mapping relationship, so as to obtain the morphological features after writing the data, which can be referred to Figure 5 .

[0113] S208. Perform feature crossing on the morphological features and the media traffic features to obtain the first crossed feature.

[0114] Among them, the media traffic feature may be a feature used to represent features related to media traffic, including at least one of a media site identifier, a scene identifier, or a media display position identifier.

[0115] In one embodiment, the server obtains the media traffic feature, performs hash processing on the morphological feature to obtain a morphological feature value; performs hash processing on the media traffic feature to obtain a media traffic feature value; and performs feature crossing on the morphological feature value and the media traffic feature value to obtain the first crossed feature.

[0116] For example, the server performs hash processing on at least one of the media site identifier, the scene identifier, or the media display position identifier to obtain at least one hash value of the media site identifier value, the scene identifier value, or the media display position identifier value; performs hash processing on the morphological feature to obtain a morphological feature value, which can be referred to Figure 5; Then, cross the morphological feature value with at least one hash value among the media site identification value, the scene identification value, or the media display bit identification value to obtain a first cross feature.

[0117] S210, through a media prediction model, perform feature processing on the first cross feature to obtain a first target feature, perform feature processing on the first cross feature, the morphological feature, the media feature of the media to be recommended, and the object feature of the target object to obtain a second target feature, and determine the interaction rate of the media to be recommended according to the first target feature and the second target feature.

[0118] Among them, the media prediction model may include a model for predicting the media click-through rate (i.e., the click-through rate prediction model) and a model for predicting the media conversion rate (i.e., the conversion rate prediction model). For example, in an advertising scenario, the click-through rate prediction model (pCTR) can predict the click probability of a user on a candidate advertisement, and the conversion rate prediction model (pCVR) can predict the conversion probability of a user on a candidate advertisement.

[0119] The click-through rate prediction model may include two network branches, which can be referred to Figure 6 . The number of network layers and the number of parameters of the first network branch in these two network branches are less than those of the second network branch. The first network branch can be a network of logistic regression, with strong memory ability; the second network branch can be a deep neural network, with strong generalization ability, which can deeply explore potential features in the data. Even for very sparse feature inputs, a relatively stable and smooth interaction probability can be obtained. Among them, the structure of the conversion rate prediction model is similar to that of the click-through rate prediction model.

[0120] When the media to be recommended is a media for promotion, the media feature of the media to be recommended can be at least one of the playing duration of the media to be recommended, the interaction mode display parameter at the corresponding playing progress, or the production party information, etc. When the media to be recommended is a media task for promotion, the media feature of the media to be recommended can be at least one of the media playing duration corresponding to the media task, the interaction mode display parameter at the corresponding playing progress, or the production party information, etc.

[0121] The object feature of the target object can be the basic information, interest information, and operation behavior on the client of the target object, etc.

[0122] In one embodiment, the server can also obtain context features, and then through the media prediction model, perform feature processing on the first cross feature to obtain a first target feature, perform feature processing on the first cross feature, the morphological feature, the media feature of the media to be recommended, the object feature of the target object, and the context features to obtain a second target feature, and determine the interaction rate of the media to be recommended according to the first target feature and the second target feature.

[0123] Among them, the context features can be time information and network information of the terminal held by the target object. Utilizing these context features can help improve the accuracy of the interaction rate.

[0124] In one embodiment, the server can perform feature processing on the first cross feature through a click-through rate prediction model to obtain a first target feature, perform feature processing on the first cross feature, morphological feature, media feature of the media to be recommended, and object feature of the target object to obtain a second target feature, and determine the click-through rate of the media to be recommended according to the first target feature and the second target feature.

[0125] For example, the server can perform feature processing on the first cross feature through the first network branch in the click-through rate prediction model to obtain a first target feature; perform feature processing on the first cross feature, morphological feature, media feature of the media to be recommended, and object feature of the target object through the second network branch in the click-through rate prediction model to obtain a second target feature, and determine the click-through rate of the media to be recommended according to the first target feature and the second target feature.

[0126] In addition, through a conversion rate prediction model, perform feature processing on the first cross feature to obtain a first target feature, perform feature processing on the first cross feature, morphological feature, media feature of the media to be recommended, and object feature of the target object to obtain a second target feature, and determine the click-through rate of the media to be recommended according to the first target feature and the second target feature. It should be noted that the first target feature and the second target feature extracted by the click-through rate prediction model may be different from the first target feature and the second target feature extracted by the conversion rate prediction model.

[0127] For example, the server can perform feature processing on the first cross feature through the first network branch in the conversion rate prediction model to obtain a first target feature; perform feature processing on the first cross feature, morphological feature, media feature of the media to be recommended, and object feature of the target object through the second network branch in the conversion rate prediction model to obtain a second target feature, and determine the conversion rate of the media to be recommended according to the first target feature and the second target feature.

[0128] When the first cross feature is a feature of the hash value type, before the server can also process to obtain the second target feature, perform hash processing on the morphological feature, media feature of the media to be recommended, and object feature of the target object, and then perform feature processing on the first cross feature and the hashed morphological feature, media feature of the media to be recommended, and object feature of the target object to obtain the second target feature. It should be noted that when processing to obtain the second target feature as described above, context features can be added for feature processing together.

[0129] In one embodiment, after predicting the interaction rate, the server may select at least one media from the media to be recommended according to the interaction rate, and then push the selected media and the corresponding media form data to the target object, so as to display the media on the application page of the terminal held by the target object according to the media form corresponding to the media form data.

[0130] In the above embodiment, before estimating the interaction rate using the media estimation model, at least two media characteristic data are parsed from the media form data of the media to be recommended, and each media characteristic data is split into different form characteristic parameters. Each form characteristic parameter of each media characteristic data is mapped into a form feature of a preset type in turn. Thus, no matter how the media form changes, it can be transformed into a form feature that can represent the media form. Therefore, when estimating the interaction rate using the media estimation model, not only can the perception of the new media form be realized, but also the estimation accuracy can be effectively improved; in addition, feature crossing is also performed on the form feature and the media traffic feature. Therefore, no matter what the source of the media traffic is, by using the media estimation model to perform feature processing on the first cross feature to obtain the first target feature, and performing feature processing on the first cross feature, the form feature, the media feature of the media to be recommended, and the object feature of the target object to obtain the second target feature, the interaction rate of the media to be recommended can be accurately determined according to the first target feature and the second target feature, further improving the accuracy of the interaction rate estimation.

[0131] In one embodiment, the media estimation model is an online media estimation model; the method further includes: the server inputs the media form data of the media to be recommended into an offline network model, so that the offline network model extracts the embedded representation of the media form data to obtain a third embedded representation; the third embedded representation is used as a form feature to perform feature crossing with the media traffic feature to obtain a second cross feature; through the online media estimation model, feature processing is performed on the second cross feature to obtain a third target feature, and feature processing is performed on the second cross feature, the third embedded representation, the media feature of the media to be recommended, and the object feature of the target object to obtain a fourth target feature, and the interaction rate of the media to be recommended is determined according to the third target feature and the fourth target feature.

[0132] Among them, for similar media form data, similar third embedded representations can be extracted. Thus, for similar media form data, the corresponding interaction rate can be accurately estimated. In the training stage, by constructing a loss function, the embedded representations of similar forms are made similar, and the overall generalization of the features is improved, thereby improving the generalization of the online media estimation model.

[0133] In the above embodiments, before estimating the interaction rate using the online media estimation model, the embedded representation of the media form data can be extracted using the offline network model first, so that the online media estimation model can perceive the entire media form using one embedded representation; in addition, the obtained third embedded representation is used as a form feature to perform feature crossing with the media traffic feature to obtain the second crossed feature; therefore, regardless of the source of the media traffic, by using the online media estimation model to process the second crossed feature to obtain the first target feature, and processing the second crossed feature, form feature, media feature of the media to be recommended, and object feature of the target object to obtain the second target feature, the interaction rate of the media to be recommended can be accurately determined according to the first target feature and the second target feature, further improving the accuracy of the interaction rate estimation.

[0134] In one embodiment, as Figure 7 shown, the method further includes:

[0135] S702, Parse the media form data of the sample media to obtain at least two training media characteristic data.

[0136] Among them, the sample media can be the media used to promote the product that needs to be pushed to the sample object. At this time, the presentation form of the sample media can be one or more (i.e., two or more data) combinations of videos, images, animations, and special effect texts about the product. The product can be intelligent devices, cars, games, application software, etc. that need to be promoted. In addition, the sample media can also be the media task used to promote the product that needs to be pushed to the sample object.

[0137] The media form data of the sample media can refer to the data used to represent the form presented by the sample media, which can be graphical or graphic-text form data. For example, each frame of media form data can be graphical or graphic-text form data used to represent the form presented by the sample media, as Figure 3 shown. Among them, the media form data in different stages can be different. For example, the media form data in the display stage, playback stage, interaction stage, and conversion stage of the sample media can be different.

[0138] The sample object can be a user who needs to be pushed the media to be recommended during the application interaction in scenarios such as historical video viewing (including on-demand videos, live videos, and videos in the live broadcast room), playing games, browsing the web, and others. For example, when a user is watching a video, in order to increase the click-through rate and conversion rate of a product that a certain user may be interested in, an advertisement for the product can be pushed to the user.

[0139] The training media characteristic data can be the data used to represent the interaction characteristics of the sample media, such as swipe click, shake click, and card click, etc.

[0140] S704. Split each piece of training media characteristic data into different training form characteristic parameters respectively.

[0141] Among them, the training form characteristic parameters can be parameters obtained by splitting the training media characteristic data of the sample media and used to represent the small-particle interaction characteristics of the media to be recommended.

[0142] S706. Map each training form characteristic parameter into a training form feature of a preset type in sequence.

[0143] Among them, there are a fixed number of types (i.e., preset types) of training form features, that is, there are a fixed number of training form features, including: component display timing, component style, media display time, playback attribute, click mode, click area, retention prompt (such as a retention pop-up window), landing page style, download mode, landing page pull-up method, and incentive mode, etc. The above training form features are only examples rather than an exhaustive list, and can be dynamically added or deleted according to the actual situation.

[0144] It should be noted that the number of training form characteristic parameters is greater than the number of media characteristic data, and the number of media characteristic data is greater than the number of training form features, that is: the relationship between training form characteristic parameters and media characteristic data is many-to-one, and the relationship between media characteristic data and training form features is many-to-one.

[0145] In one embodiment, when new media form data appears, the training form characteristic parameters corresponding to the new media form data can be mapped to the training form features corresponding to similar form characteristics, so as to obtain the mapped training form features.

[0146] S708. Perform feature crossing on the training form features and the media traffic features corresponding to the sample media to obtain a third crossed feature.

[0147] Among them, the media traffic features corresponding to the sample media can be features used to represent features related to media traffic, including at least one of media site identifier, scene identifier, or media display position identifier.

[0148] S710. Through the media prediction model, perform feature processing on the third crossed feature to obtain a third target feature, perform feature processing on the third crossed feature, training form features, media features of the sample media, and object features of the sample object to obtain a fourth target feature, and determine the interaction rate of the sample media according to the third target feature and the fourth target feature.

[0149] Among them, when the sample media is the media for promotion, the media features of the sample media may be at least one of the playing duration of the sample media, the interaction mode display parameter at the corresponding playing progress, or the production party information, etc. When the sample media is a media task for promotion, the media features of the sample media may be at least one of the media playing duration corresponding to the media task, the interaction mode display parameter at the corresponding playing progress, or the production party information, etc.

[0150] The object features of the sample object may be the basic information, interest information, and operation behaviors on the client of the sample object, etc.

[0151] It should be noted that the detailed steps of the above S702 - S710 can be referred to Figure 2 Examples.

[0152] In one embodiment, the server can also obtain the context features for training (abbreviated as training context features), and then through the media prediction model, perform feature processing on the second cross - feature to obtain the first target feature, perform feature processing on the second cross - feature, morphological features, media features of the media to be recommended, object features of the target object, and context features to obtain the second target feature, and determine the interaction rate of the media to be recommended according to the first target feature and the second target feature.

[0153] Among them, the context features may be time information and network information of the terminal held by the target object. Utilizing the training context features can help improve the generalization ability of the media prediction model, thereby helping to improve the accuracy of the interaction rate.

[0154] S712, optimize the media prediction model according to the loss value between the interaction rate and the interaction label of the sample media.

[0155] Among them, the interaction label can be used to represent the interaction situation of whether the sample object interacts with the sample media, such as whether the sample media is clicked, and whether there is conversion of the product corresponding to the sample media, such as whether it is downloaded or the game of the advertisement is experienced.

[0156] In one embodiment, the server can determine the loss value between the interaction rate and the interaction label of the sample media, and then perform backpropagation to calculate the parameter gradient values of each network layer in the media prediction model, and optimize the parameters of each network layer in the media prediction model according to the parameter gradient values until the model converges and stops training. It should be noted that the media prediction model can be an online media prediction model, so when training in the above - mentioned manner, it can be online training.

[0157] In the above embodiments, at least two training media characteristic data are parsed from the media form data of the sample media. Each training media characteristic data is split into different training form characteristic parameters, and the training form characteristic parameters of each training media characteristic data are mapped into training form features of a preset type in sequence. Thus, no matter how the media form changes, it can be converted into training form features that can represent the media form. Therefore, when using the media prediction model to predict the interaction rate, not only can the perception of the new media form be realized, the efficiency of inputting the form features into the media prediction model be greatly improved, but also the performance overhead of the media prediction model can be significantly reduced. In addition, feature crossing is performed on the training form features and the media traffic features. Therefore, no matter what the source of the media traffic is, by using the media prediction model to perform feature processing on the first cross feature to obtain the first target feature, and performing feature processing on the first cross feature, the training form features, the media features of the sample media, and the object features of the sample object to obtain the second target feature, the interaction rate of the sample media can be predicted based on the first target feature and the second target feature, and the media prediction model can be optimized according to the loss value between the interaction rate and the interaction label of the sample media. Therefore, the accuracy of the media prediction model for predicting the interaction rate can be effectively improved.

[0158] In one embodiment, the media data processing method of the present application can be applied to a recommendation system, such as a recommendation system installed on application programs such as social applications, video applications, and game applications. Therefore, after S210, the server can use the interaction rate to select corresponding media for recommendation through the recommendation system. Considering that the developer or service provider of the recommendation system can recommend its own (i.e., the party's) products through the media, or can recommend the products of the product party (i.e., the product developer, which can be called the other party) through the media, the following application scenarios can be described:

[0159] Scenario 1: The scenario of recommending the party's own products.

[0160] In one embodiment, the interaction rate includes the conversion rate and the click-through rate, and the number of media to be recommended is at least two; the method further includes: the server selects a first media whose interaction rate meets the interaction condition from at least two media to be recommended; and pushes the first media and the media form data corresponding to the first media to the target object, so as to display the first media on the application page of the target object according to the media form data.

[0161] When pushing the media, the server can also first generate a target recommended media according to the first media and the media form data matched with the first media, and then push the target recommended media to the target object, so as to display the target recommended media on the application page of the target object.

[0162] Scenario 2: The scenario of recommending the products of the other party.

[0163] In one embodiment, the interaction rate includes the conversion rate and the click-through rate, and the number of media to be recommended is at least two; the method further includes: the server can obtain the media recommendation resource value proposed by the product party for recommending the product of the product party; among at least two media to be recommended, select the second media according to the media recommendation resource value, the conversion rate, and the click-through rate; push the second media and the media form data corresponding to the second media to the target object, so as to display the second media on the application page of the target object according to the media form data.

[0164] Among them, the media recommendation resource value may be the digital resource value that the product party has to spend when promoting the product. For example, in order to increase the download volume of a newly released software or increase the sales volume of the products sold, a certain product party can compete with other product parties. Therefore, in this scenario, the media recommendation resource value may refer to the price of the competition.

[0165] In one embodiment, after obtaining the media recommendation resource value, the server calculates the display resource value under the target unit quantity according to the media recommendation resource value, the conversion rate, and the click-through rate, and then selects the second media from at least two media to be recommended according to the display resource value, and then pushes the second media and the media form data corresponding to the second media to the target object, so as to display the second media on the application page of the terminal held by the target object according to the media form data; when the second media is a media task, the media data corresponding to the second media can be displayed on the application page of the terminal held by the target object according to the media form data; among them, the media data may be the media playback data corresponding to the media task, such as an advertisement.

[0166] Among them, the display resource value may be the resource value that the developer or service provider of the recommendation system can obtain for every thousand times of displaying the media. For example, in the advertisement scenario, the display resource value may be the expected revenue per thousand impressions (eCPM), that is, the expected advertisement revenue obtained for every thousand times of displaying the advertisement.

[0167] In one embodiment, the server can also first generate a target recommended media according to the second media and the media form data matched with the second media, and then push the target recommended media to the target object, so as to display the target recommended media on the application page of the target object.

[0168] In the above embodiment, using the conversion rate to select the most suitable media to push to the target object can effectively improve the click-through rate and conversion rate of the product corresponding to the media.

[0169] As an example, in order to more clearly understand the solution of the present application, here, in combination with the advertisement scenario, the click-through rate prediction model, the conversion rate prediction model, Figures 3 - 6 and Figures 8 - 14 it is described as follows:

[0170] The advertising form feature framework proposed in this application is as follows: form configuration platform, Mixer (advertising form matching), Ranking Server (form feature construction), and training of the pCXR model (optimizing the pCXR model by introducing form features), as Figure 8 shown. Among them, the pCXR model generally refers to the click-through rate prediction model and the conversion rate prediction model.

[0171] Next, the form configuration platform, advertising form matching, form feature construction, and training of the pCXR model will be introduced as follows:

[0172] a) Form configuration platform.

[0173] The form configuration platform is a design platform for advertising forms. Based on the product, it designs appropriate advertising display forms for corresponding traffic, users, or advertisements, and finally takes effect on the front end as the advertising request is sent down, becoming the display advertisement seen by users.

[0174] From the advertisement being exposed to users to the final conversion, it can be divided into 4 stages: advertisement display, advertisement playback, advertisement interaction, and advertisement conversion. Among them, the advertisement display forms in the advertisement display stage are as Figure 9 and Figure 10 shown. Figure 9 is the incentive dynamic card form, Figure 10 is the immersive card form, and the displayed styles can be different at different times; the advertisement display forms in the advertisement playback stage are as Figure 11 shown. The advertisement playback (i.e., the opening method) includes auto-play and voice control playback; the advertisement display forms in the advertisement interaction stage are as Figure 12 shown; the advertisement display forms in the advertisement conversion are as Figure 13 shown. For the advertisement forms in each stage, they can be designed on the form configuration platform to create various advertisement forms.

[0175] b) Advertising form matching.

[0176] After the user request arrives, the backend framework Mixer matches the most suitable advertising form for each advertisement in the advertisement candidate queue. If an advertisement wins, its corresponding advertising form will also be sent down to the front end for display and take effect.

[0177] c) Form feature construction.

[0178] The system defines 11 form features, representing all 11 functions of the advertising form in the entire advertising life cycle. Then, the display form of each advertisement in the candidate queue will be disassembled into a series of form characteristics in form feature construction, and each form characteristic will be further disassembled into multiple characteristic parameters, such as Figure 14 andFigure 15 As shown. Among them, each morphological characteristic is mapped to the corresponding morphological feature according to the system configuration.

[0179] For the specific mapping process, refer to Figure 5 . All the characteristic parameters under each morphological characteristic in the advertisement display form will be written into the specified morphological feature PB (Protocol Buffers). One morphological feature may correspond to multiple different morphological characteristic parameters.

[0180] Each morphological feature generates its corresponding morphological feature value through a hash operation, and thus the morphological feature of a candidate advertisement is generated. The overall process can be referred to Figure 5 . When a new morphological characteristic or characteristic parameter needs to be configured for a certain advertisement display form, only the mapping relationship between the parameter corresponding morphological characteristic and the morphological feature needs to be configured, so that it can be automatically written into the existing morphological feature PB. Finally, a brand-new morphological feature value is calculated through the Hash operation, representing the overall CXR performance of this new form under the advertisement feature.

[0181] d) Training of the pCXR model.

[0182] After obtaining the morphological feature, send this morphological feature into the pCXR model to participate in the model prediction.

[0183] Considering that the CXR performance of the same form is different in different traffic, cross the morphological feature with the advertisement traffic feature, which can be referred to Figure 6 . Then send the crossed morphological feature into both the Wide and Deep sides of the pCXR model at the same time to accelerate the model convergence.

[0184] In addition to obtaining the morphological feature in the way of step c) above, a unique embedding representation can also be learned for each advertisement display form through an offline model, and use this embedding as the morphological feature to input into the pCXR model. On the one hand, it can enable the pCXR model to perceive the entire form with only one feature; on the other hand, a loss function can be constructed to make the embeddings of similar forms similar, improving the generalization of the pCXR model.

[0185] Through the solutions of the above embodiments, the following beneficial effects can be achieved:

[0186] 1) The model perception of a large number of advertisement display form parameters is completed with extremely low performance overhead.

[0187] 2) Realize the automatic and real-time perception of the new form model, reduce the pCXR Bias brought by the form to less than 5%, and greatly improve the iteration efficiency of the advertisement display form.

[0188] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise clearly stated in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.

[0189] Based on the same inventive concept, an embodiment of the present application also provides a media data processing device for implementing the above-mentioned media data processing method. The solution provided by this device to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the media data processing device provided below can refer to the limitations on the media data processing method in the above text, and will not be repeated here.

[0190] In one embodiment, as Figure 16 shown, a media data processing device is provided, including: a parsing module 1602, a splitting module 1604, a mapping module 1606, a cross module 1608, and an estimation module 1610, where:

[0191] The parsing module 1602 is configured to parse at least two media characteristic data from the media form data of the media to be recommended;

[0192] The splitting module 1604 is configured to split each media characteristic data into different form characteristic parameters respectively;

[0193] The mapping module 1606 is configured to map the form characteristic parameters of each media characteristic data into form characteristics of a preset type in sequence;

[0194] The cross module 1608 is configured to perform feature crossing on the form characteristics and the media traffic characteristics to obtain a first cross feature;

[0195] The estimation module 1610 is configured to perform feature processing on the first cross feature through a media estimation model to obtain a first target feature, perform feature processing on the first cross feature, the form characteristics, the media characteristics of the media to be recommended, and the object characteristics of the target object to obtain a second target feature, and determine the interaction rate of the media to be recommended according to the first target feature and the second target feature.

[0196] In one of the embodiments, the number of media to be recommended is at least two; the device further includes:

[0197] A receiving module, configured to receive a media recommendation request sent by a client;

[0198] A first obtaining module, configured to obtain at least two media to be recommended based on the media recommendation request;

[0199] A storage module, configured to save at least two media to be recommended in a media candidate queue;

[0200] A searching module, configured to sequentially find matching media form data for each media to be recommended in the media candidate queue.

[0201] In one embodiment, the searching module is further configured to obtain media form selection conditions configured on a media form configuration platform; according to the media form selection conditions, select media form data matching each media to be recommended in the candidate media form data; wherein, the media form configuration platform is used to create various types of candidate media form data.

[0202] In one embodiment, the apparatus further includes:

[0203] The first obtaining module is further configured to obtain media form description data of an operation object on a media form configuration platform;

[0204] A first generating module, configured to generate candidate media form data in an image or text - image form according to the media form description data, media traffic characteristics, and object characteristics of a target object;

[0205] The storage module is further configured to, in response to a data storage operation triggered on the media form configuration platform, store the candidate media form data in a database;

[0206] The first obtaining module is further configured to obtain at least two media to be recommended corresponding to the media recommendation request from the database.

[0207] In one embodiment, the mapping module is further configured to use the parameter names in the form characteristic parameters of each media characteristic as keys, and the parameter values in the form characteristic parameters of each media characteristic as values, to obtain a plurality of different key - value pairs; obtain the mapping relationship between the form characteristic parameters and the form characteristics of a preset type; according to the mapping relationship, write each key - value pair into a form characteristic file in a target format to obtain the form characteristics of the preset type.

[0208] In one embodiment, the apparatus further includes:

[0209] A second obtaining module, configured to obtain media traffic characteristics; the media traffic characteristics include at least one of a media site identifier, a scene identifier, or a media display position identifier;

[0210] The cross module is also used to perform hashing processing on the morphological features to obtain morphological feature values;

[0211] Perform hashing processing on the media traffic features to obtain media traffic feature values; perform feature crossing on the morphological feature values and the media traffic feature values to obtain a first cross feature.

[0212] In one of the embodiments, the device further includes:

[0213] A third acquisition module, configured to acquire the media features of the media to be recommended and the object features of the target object;

[0214] The estimation module is also used to encode the first cross feature through the first network branch in the media estimation model to obtain a first embedding representation, and perform feature processing on the first embedding representation to obtain a first target feature; through the second network branch in the media estimation model, encode the first cross feature, the morphological features, the media features of the media to be recommended, and the object features of the target object to obtain a second embedding representation, and perform feature processing on the second embedding representation to obtain a second target feature.

[0215] In the above embodiments, before estimating the interaction rate using the media estimation model, at least two media characteristic data are parsed from the media morphological data of the media to be recommended, each media characteristic data is split into different morphological characteristic parameters, and the morphological characteristic parameters of each media characteristic data are sequentially mapped into morphological features of a preset type, so that no matter how the media form changes, it can be transformed into morphological features that can represent the inner body form. Therefore, when estimating the interaction rate using the media estimation model, not only can the perception of the new media form be realized, but also the estimation accuracy can be effectively improved; in addition, feature crossing is also performed on the morphological features and the media traffic features. Therefore, no matter what the source of the media traffic is, using the media estimation model to perform feature processing on the first cross feature to obtain a first target feature, and performing feature processing on the first cross feature, the morphological features, the media features of the media to be recommended, and the object features of the target object to obtain a second target feature, the interaction rate of the media to be recommended can be accurately determined according to the first target feature and the second target feature, further improving the accuracy of the interaction rate estimation.

[0216] In one of the embodiments, the media estimation model is an online media estimation model; the device further includes:

[0217] An extraction module, configured to input the media morphological data of the media to be recommended into an offline network model, so that the offline network model extracts the embedding representation of the media morphological data to obtain a third embedding representation;

[0218] The cross module is also used to use the third embedding representation as the morphological feature to perform feature crossing with the media traffic feature to obtain a second cross feature;

[0219] The prediction module is further configured to perform feature processing on the second cross feature through an online media prediction model to obtain a third target feature, perform feature processing on the second cross feature, the third embedding representation, the media feature of the media to be recommended, and the object feature of the target object to obtain a fourth target feature, and determine the interaction rate of the media to be recommended according to the third target feature and the fourth target feature.

[0220] In the above embodiment, before predicting the interaction rate using the online media prediction model, the embedding representation of the media form data can be extracted using an offline network model first, so that an embedding representation can be used to enable the online media prediction model to perceive the entire media form; in addition, the obtained third embedding representation is used as a form feature to perform feature crossing with the media traffic feature to obtain a second cross feature; therefore, regardless of the source of the media traffic, using the first target feature obtained by performing feature processing on the second cross feature through the online media prediction model, and the second target feature obtained by performing feature processing on the second cross feature, the form feature, the media feature of the media to be recommended, and the object feature of the target object, the interaction rate of the media to be recommended can be accurately determined according to the first target feature and the second target feature, further improving the accuracy of the interaction rate prediction.

[0221] In one of the embodiments, the apparatus further includes:

[0222] The parsing module is further configured to parse the media form data of the sample media to obtain at least two training media characteristic data;

[0223] The splitting module is further configured to split each training media characteristic data into different training form characteristic parameters respectively;

[0224] The mapping module is further configured to map each training form characteristic parameter into a training form feature of a preset type in sequence;

[0225] The crossing module is further configured to perform feature crossing on the training form feature and the media traffic feature corresponding to the sample media to obtain a third cross feature;

[0226] The prediction module is further configured to perform feature processing on the third cross feature through a media prediction model to obtain a third target feature, perform feature processing on the third cross feature, the training form feature, the media feature of the sample media, and the object feature of the sample object to obtain a fourth target feature, and determine the interaction rate of the sample media according to the third target feature and the fourth target feature;

[0227] The optimization module is configured to optimize the media prediction model according to the loss value between the interaction rate and the interaction label of the sample media.

[0228] In the above embodiments, at least two training media characteristic data are parsed from the media form data of the sample media. Each training media characteristic data is split into different training form characteristic parameters, and the training form characteristic parameters of each training media characteristic data are mapped into training form features of a preset type in sequence. Thus, no matter how the media form changes, it can be converted into training form features that can represent the media form. Therefore, when using the media prediction model to predict the interaction rate, not only can the perception of the new media form be realized, the efficiency of inputting form features into the media prediction model be greatly improved, but also the performance overhead of the media prediction model can be significantly reduced. In addition, feature crossing is performed on the training form features and media traffic features. Therefore, no matter what the source of the media traffic is, by using the media prediction model to perform feature processing on the first cross feature to obtain the first target feature, and performing feature processing on the first cross feature, training form features, media features of the sample media, and object features of the sample object to obtain the second target feature, the interaction rate of the sample media can be predicted based on the first target feature and the second target feature, and the media prediction model can be optimized according to the loss value between the interaction rate and the interaction label of the sample media. Therefore, the accuracy of the media prediction model for predicting the interaction rate can be effectively improved.

[0229] In one of the embodiments, the interaction rate includes the conversion rate and the click-through rate, and the number of media to be recommended is at least two; the apparatus further includes:

[0230] A fourth acquisition module, configured to acquire a media recommendation resource value proposed by the product party for recommending the product of the product party; and select a second media from at least two media to be recommended according to the media recommendation resource value, conversion rate, and click-through rate.

[0231] A first push module, configured to push the second media and the media form data corresponding to the second media to the target object, so as to display the second media on the application page of the target object according to the media form data.

[0232] In one of the embodiments, the interaction rate includes the conversion rate and the click-through rate, and the number of media to be recommended is at least two; the apparatus further includes:

[0233] A selection module, configured to select a first media whose interaction rate meets the interaction condition from at least two media to be recommended.

[0234] A second push module, configured to push the first media and the media form data corresponding to the first media to the target object, so as to display the first media on the application page of the target object according to the media form data.

[0235] In the above embodiments, by using the conversion rate to select the most suitable media to push to the target object, the click-through rate and conversion rate of the product corresponding to the media can be effectively improved.

[0236] Each module in the above media data processing device can be implemented in whole or in part by software, hardware, or a combination thereof. Each of the above modules can be embedded in the processor of the computer device in hardware form or be independent of it, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each of the above modules.

[0237] In one embodiment, a computer device is provided. The computer device can be a server, and its internal structure diagram can be as Figure 17 shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store media data. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, it implements a media data processing method.

[0238] Those skilled in the art can understand that Figure 17 the structure shown in

[0239] is only a block diagram of a part of the structure related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0240] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the above media data processing method.

[0241] In one embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, it implements the steps of the above media data processing method.

[0242] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data that have been authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with the relevant laws, regulations, and standards of relevant countries and regions.

[0243] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in this application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in this application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0244] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0245] The above-described embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation to the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all fall within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A media data processing method, characterized in that, The method includes: Parsing at least two media characteristic data from the media form data of the media to be recommended; Respectively splitting each of the media characteristic data into different form characteristic parameters; Sequentially mapping the form characteristic parameters of each of the media characteristic data into form features of a preset type; Performing feature crossing on the form features and media traffic features to obtain first cross features; Through a media prediction model, performing feature processing on the first cross features to obtain first target features, performing feature processing on the first cross features, the form features, the media features of the media to be recommended, and the object features of the target object to obtain second target features, and determining the interaction rate of the media to be recommended according to the first target features and the second target features.

2. The method according to claim 1, wherein The number of the media to be recommended is at least two; before parsing at least two media characteristic data from the media form of the media to be recommended, the method further includes: Receiving a media recommendation request sent by a client; Obtaining at least two media to be recommended based on the media recommendation request; Saving at least two of the media to be recommended in a media candidate queue; Sequentially finding matching media form data for each of the media to be recommended in the media candidate queue.

3. The method according to claim 2, wherein The sequentially finding matching media form data for each of the media to be recommended in the media candidate queue includes: Obtaining the media form selection conditions configured on a media form configuration platform; Selecting, according to the media form selection conditions, media form data matching each of the media to be recommended in the candidate media form data; Wherein, the media form configuration platform is used to create various types of candidate media form data.

4. The method according to claim 3, wherein The method further includes: Obtaining media form description data of an operation object on the media form configuration platform; Generating candidate media form data in an image or text-and-graphic form according to the media form description data, the media traffic features, and the object features of the target object; In response to a data storage operation triggered on the media form configuration platform, storing the candidate media form data in a database; The obtaining at least two media to be recommended based on the media recommendation request includes: Obtaining at least two media to be recommended corresponding to the media recommendation request from the database.

5. The method according to claim 1, characterized in that, The sequentially mapping the form characteristic parameters of each of the media characteristic data into form features of a preset type includes: Using the parameter names in the form characteristic parameters of each of the media characteristics as keys and the parameter values in the form characteristic parameters of each of the media characteristics as values to obtain a plurality of different key-value pairs; Obtaining the mapping relationship between the form characteristic parameters and the form features of the preset type; Writing each of the key-value pairs into a form feature file in a target format according to the mapping relationship to obtain form features of a preset type.

6. The method according to claim 1, wherein The method further includes: Obtaining media traffic features; the media traffic features include at least one of a media site identifier, a scenario identifier, or a media display position identifier; The performing feature crossing on the form features and media traffic features to obtain first cross features includes: Performing hash processing on the form features to obtain form feature values; Hash the media traffic characteristics to obtain media traffic characteristic values; Perform feature crossing on the morphological characteristic values and the media traffic characteristic values to obtain first crossed features.

7. The method according to claim 1, characterized in that, The method further includes: Obtain the media characteristics of the media to be recommended and the object characteristics of the target object; The performing, by a media prediction model, feature processing on the first crossed features to obtain first target features, and performing feature processing on the first crossed features, the morphological characteristics, the media characteristics of the media to be recommended, and the object characteristics of the target object to obtain second target features includes: Encoding the first crossed features through a first network branch in the media prediction model to obtain a first embedding representation, and performing feature processing on the first embedding representation to obtain first target features; Encoding the first crossed features, the morphological characteristics, the media characteristics of the media to be recommended, and the object characteristics of the target object through a second network branch in the media prediction model to obtain a second embedding representation, and performing feature processing on the second embedding representation to obtain second target features.

8. The method according to claim 1, characterized in that, The media prediction model is an online media prediction model; the method further includes: Input the media morphological data of the media to be recommended into an offline network model, so that the offline network model extracts the embedding representation of the media morphological data to obtain a third embedding representation; Use the third embedding representation as morphological characteristics to perform feature crossing with media traffic characteristics to obtain second crossed features; Perform feature processing on the second crossed features through the online media prediction model to obtain third target features, perform feature processing on the second crossed features, the third embedding representation, the media characteristics of the media to be recommended, and the object characteristics of the target object to obtain fourth target features, and determine the interaction rate of the media to be recommended according to the third target features and the fourth target features.

9. The method according to any one of claims 1 to 7, characterized in that The method further includes: Parse the media morphological data of the sample media to obtain at least two training media characteristic data; Respectively split each of the training media characteristic data into different training morphological characteristic parameters; Successively map each of the training morphological characteristic parameters into training morphological characteristics of the preset type; Perform feature crossing on the training morphological characteristics and the media traffic characteristics corresponding to the sample media to obtain third crossed features; Perform feature processing on the third crossed features through a media prediction model to obtain third target features, perform feature processing on the third crossed features, the training morphological characteristics, the media characteristics of the sample media, and the object characteristics of the sample object to obtain fourth target features, and determine the interaction rate of the sample media according to the third target features and the fourth target features; Optimize the media prediction model according to the loss value between the interaction rate and the interaction label of the sample media.

10. The method according to any one of claims 1 to 7, characterized in that, The interaction rate includes a conversion rate and a click-through rate, and the number of media to be recommended is at least two; the method further includes: Among at least two of the media to be recommended, select a first media whose interaction rate meets the interaction condition; Push the first media and the media form data corresponding to the first media to the target object, so as to display the first media on the application page of the target object according to the media form data.

11. The method according to any one of claims 1 to 7, characterized in that, The interaction rate includes the conversion rate and the click-through rate, and the number of media to be recommended is at least two; the method further includes: Obtain the media recommendation resource value proposed by the product party for recommending the product of the product party. Select a second media from at least two of the media to be recommended according to the media recommendation resource value, the conversion rate and the click-through rate. Push the second media and the media form data corresponding to the second media to the target object, so as to display the second media on the application page of the target object according to the media form data.

12. A media data processing device, characterized in that, The device includes: A parsing module, configured to parse at least two media characteristic data from the media form data of the media to be recommended. A splitting module, configured to split each of the media characteristic data into different form characteristic parameters. A mapping module, configured to sequentially map the form characteristic parameters of each of the media characteristic data into form features of a preset type. A cross module, configured to perform feature crossing on the form features and the media traffic features to obtain a first cross feature. An estimation module, configured to perform feature processing on the first cross feature through a media estimation model to obtain a first target feature, perform feature processing on the first cross feature, the form features, the media features of the media to be recommended, and the object features of the target object to obtain a second target feature, and determine the interaction rate of the media to be recommended according to the first target feature and the second target feature.

13. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, the steps of the method according to any one of claims 1 to 11 are implemented.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 11 are implemented.

15. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 11 are implemented.

Images